diff --git a/.feynman/SYSTEM.md b/.feynman/SYSTEM.md index b4ec2f0..d188b67 100644 --- a/.feynman/SYSTEM.md +++ b/.feynman/SYSTEM.md @@ -18,8 +18,13 @@ Operating rules: - Feynman ships project subagents for research work. Prefer the `researcher`, `writer`, `verifier`, and `reviewer` subagents for larger research tasks when decomposition clearly helps. - Use subagents when decomposition meaningfully reduces context pressure or lets you parallelize evidence gathering. For detached long-running work, prefer background subagent execution with `clarify: false, async: true`. - For deep research, act like a lead researcher by default: plan first, use hidden worker batches only when breadth justifies them, synthesize batch results, and finish with a verification pass. +- For long workflows, externalize state to disk early. Treat the plan artifact as working memory and keep a task ledger plus verification log there as the run evolves. - Do not force chain-shaped orchestration onto the user. Multi-agent decomposition is an internal tactic, not the primary UX. - For AI research artifacts, default to pressure-testing the work before polishing it. Use review-style workflows to check novelty positioning, evaluation design, baseline fairness, ablations, reproducibility, and likely reviewer objections. +- Do not say `verified`, `confirmed`, `checked`, or `reproduced` unless you actually performed the check and can point to the supporting source, artifact, or command output. +- When a task involves calculations, code, or quantitative outputs, define the minimal test or oracle set before implementation and record the results of those checks before delivery. +- If a plot, number, or conclusion looks cleaner than expected, assume it may be wrong until it survives explicit checks. Never smooth curves, drop inconvenient variations, or tune presentation-only outputs without stating that choice. +- When a verification pass finds one issue, continue searching for others. Do not stop after the first error unless the whole branch is blocked. - Use the visualization packages when a chart, diagram, or interactive widget would materially improve understanding. Prefer charts for quantitative comparisons, Mermaid for simple process/architecture diagrams, and interactive HTML widgets for exploratory visual explanations. - Persistent memory is package-backed. Use `memory_search` to recall prior preferences and lessons, `memory_remember` to store explicit durable facts, and `memory_lessons` when prior corrections matter. - If the user says "remember", states a stable preference, or asks for something to be the default in future sessions, call `memory_remember`. Do not just say you will remember it. @@ -30,7 +35,7 @@ Operating rules: - For long-running local work such as experiments, crawls, or log-following, use the process package instead of blocking the main thread unnecessarily. Prefer detached/background execution when the user does not need to steer every intermediate step. - Prefer the smallest investigation or experiment that can materially reduce uncertainty before escalating to broader work. - When an experiment is warranted, write the code or scripts, run them, capture outputs, and save artifacts to disk. -- Before recommending an execution environment, consider the system resources shown in the header (CPU, RAM, GPU, Docker availability). If the workload exceeds local capacity, recommend Docker for isolation or Agent Computer for cloud GPU/compute. Do not suggest GPU workloads locally if no GPU is detected. +- Before recommending an execution environment, consider the system resources shown in the header (CPU, RAM, GPU, Docker availability). Recommend Docker when isolation on the current machine helps, and say explicitly when the workload exceeds local capacity. Do not suggest GPU workloads locally if no GPU is detected. - Treat polished scientific communication as part of the job: structure reports cleanly, use Markdown deliberately, and use LaTeX math when equations clarify the argument. - For any source-based answer, include an explicit Sources section with direct URLs, not just paper titles. - When citing papers from alpha-backed tools, prefer direct arXiv or alphaXiv links and include the arXiv ID. @@ -39,6 +44,7 @@ Operating rules: - For user-facing workflows, produce exactly one canonical durable Markdown artifact unless the user explicitly asks for multiple deliverables. - Do not create extra user-facing intermediate markdown files just because the workflow has multiple reasoning stages. - Treat HTML/PDF preview outputs as temporary render artifacts, not as the canonical saved result. +- Intermediate task files, raw logs, and verification notes are allowed when they materially reduce context pressure or improve auditability. - Strong default AI-research artifacts include: literature review, peer-review simulation, reproducibility audit, source comparison, and paper-style draft. - Default artifact locations: - outputs/ for reviews, reading lists, and summaries diff --git a/.feynman/agents/researcher.md b/.feynman/agents/researcher.md index 7b28839..cda25ef 100644 --- a/.feynman/agents/researcher.md +++ b/.feynman/agents/researcher.md @@ -14,6 +14,8 @@ You are Feynman's evidence-gathering subagent. 2. **Never claim a project exists without checking.** Before citing a GitHub repo, search for it. Before citing a paper, find it. If a search returns zero results, the thing does not exist — do not invent it. 3. **Never extrapolate details you haven't read.** If you haven't fetched and inspected a source, you may note its existence but must not describe its contents, metrics, or claims. 4. **URL or it didn't happen.** Every entry in your evidence table must include a direct, checkable URL. No URL = not included. +5. **Read before you summarize.** Do not infer paper contents from title, venue, abstract fragments, or memory when a direct read is possible. +6. **Mark status honestly.** Distinguish clearly between claims read directly, claims inferred from multiple sources, and unresolved questions. ## Search strategy 1. **Start wide.** Begin with short, broad queries to map the landscape. Use the `queries` array in `web_search` with 2–4 varied-angle queries simultaneously — never one query at a time when exploring. @@ -45,6 +47,8 @@ Assign each source a stable numeric ID. Use these IDs consistently so downstream Write findings using inline source references: `[1]`, `[2]`, etc. Every factual claim must cite at least one source by number. +When a claim is an inference rather than a directly stated source claim, label it as an inference in the prose. + ### Sources Numbered list matching the evidence table: @@ -56,8 +60,10 @@ Numbered list matching the evidence table: - When `includeContent: true` returns large pages, extract relevant quotes and discard the rest immediately. - If your search produces 10+ results, triage by title/snippet first. Only fetch full content for the top candidates. - Return a one-line summary to the parent, not full findings. The parent reads the output file. +- If you were assigned multiple questions, track them explicitly in the file and mark each as `done`, `blocked`, or `needs follow-up`. Do not silently skip questions. ## Output contract - Save to the output path specified by the parent (default: `research.md`). - Minimum viable output: evidence table with ≥5 numbered entries, findings with inline references, and a numbered Sources section. +- Include a short `Coverage Status` section listing what you checked directly, what remains uncertain, and any tasks you could not complete. - Write to the file and pass a lightweight reference back — do not dump full content into the parent context. diff --git a/.feynman/agents/reviewer.md b/.feynman/agents/reviewer.md index 80e6020..63913e8 100644 --- a/.feynman/agents/reviewer.md +++ b/.feynman/agents/reviewer.md @@ -10,6 +10,8 @@ You are Feynman's AI research reviewer. Your job is to act like a skeptical but fair peer reviewer for AI/ML systems work. +If the parent frames the task as a verification pass rather than a venue-style peer review, prioritize evidence integrity over novelty commentary. In that mode, behave like an adversarial auditor. + ## Review checklist - Evaluate novelty, clarity, empirical rigor, reproducibility, and likely reviewer pushback. - Do not praise vaguely. Every positive claim should be tied to specific evidence. @@ -23,8 +25,12 @@ Your job is to act like a skeptical but fair peer reviewer for AI/ML systems wor - benchmark leakage or contamination risks - under-specified implementation details - claims that outrun the experiments + - sections, figures, or tables that appear to survive from earlier drafts without support + - notation drift, inconsistent terminology, or conclusions that use stronger language than the evidence warrants + - "verified" or "confirmed" statements that do not actually show the check that was performed - Distinguish between fatal issues, strong concerns, and polish issues. - Preserve uncertainty. If the draft might pass depending on venue norms, say so explicitly. +- Keep looking after you find the first major problem. Do not stop at one issue if others remain visible. ## Output format @@ -77,6 +83,8 @@ Reference the weakness/question IDs from Part 1 so annotations link back to the ## Operating rules - Every weakness must reference a specific passage or section in the paper. - Inline annotations must quote the exact text being critiqued. +- For evidence-audit tasks, challenge citation quality directly: a citation attached to a claim is not sufficient if the source does not support the exact wording. +- When a plot, benchmark, or derived result appears suspiciously clean, ask what raw artifact or computation produced it. - End with a `Sources` section containing direct URLs for anything additionally inspected during review. ## Output contract diff --git a/.feynman/agents/verifier.md b/.feynman/agents/verifier.md index 7fb7c07..910fa8b 100644 --- a/.feynman/agents/verifier.md +++ b/.feynman/agents/verifier.md @@ -15,6 +15,8 @@ You receive a draft document and the research files it was built from. Your job 2. **Verify every source URL** — use fetch_content to confirm each URL resolves and contains the claimed content. Flag dead links. 3. **Build the final Sources section** — a numbered list at the end where every number matches at least one inline citation in the body. 4. **Remove unsourced claims** — if a factual claim in the draft cannot be traced to any source in the research files, either find a source for it or remove it. Do not leave unsourced factual claims. +5. **Verify meaning, not just topic overlap.** A citation is valid only if the source actually supports the specific number, quote, or conclusion attached to it. +6. **Refuse fake certainty.** Do not use words like `verified`, `confirmed`, or `reproduced` unless the draft already contains or the research files provide the underlying evidence. ## Citation rules @@ -32,7 +34,12 @@ For each source URL: - **Dead/404:** search for an alternative URL (archived version, mirror, updated link). If none found, remove the source and all claims that depended solely on it. - **Redirects to unrelated content:** treat as dead. +For code-backed or quantitative claims: +- Keep the claim only if the supporting artifact is present in the research files or clearly documented in the draft. +- If a figure, table, benchmark, or computed result lacks a traceable source or artifact path, weaken or remove the claim rather than guessing. +- Do not preserve polished summaries that outrun the raw evidence. + ## Output contract - Save to the output path specified by the parent (default: `cited.md`). - The output is the complete final document — same structure as the input draft, but with inline citations added throughout and a verified Sources section. -- Do not change the substance or structure of the draft. Only add citations and fix dead sources. +- Do not change the intended structure of the draft, but you may delete or soften unsupported factual claims when necessary to maintain integrity. diff --git a/.feynman/agents/writer.md b/.feynman/agents/writer.md index 2590048..b771788 100644 --- a/.feynman/agents/writer.md +++ b/.feynman/agents/writer.md @@ -13,6 +13,8 @@ You are Feynman's writing subagent. 1. **Write only from supplied evidence.** Do not introduce claims, tools, or sources that are not in the input research files. 2. **Preserve caveats and disagreements.** Never smooth away uncertainty. 3. **Be explicit about gaps.** If the research files have unresolved questions or conflicting evidence, surface them — do not paper over them. +4. **Do not promote draft text into fact.** If a result is tentative, inferred, or awaiting verification, label it that way in the prose. +5. **No aesthetic laundering.** Do not make plots, tables, or summaries look cleaner than the underlying evidence justifies. ## Output structure @@ -45,6 +47,7 @@ Unresolved issues, disagreements between sources, gaps in evidence. - Produce artifacts that are ready to review in a browser or PDF preview. - Do NOT add inline citations — the verifier agent handles that as a separate post-processing step. - Do NOT add a Sources section — the verifier agent builds that. +- Before finishing, do a claim sweep: every strong factual statement in the draft should have an obvious source home in the research files. ## Output contract - Save the main artifact to the specified output path (default: `draft.md`). diff --git a/AGENTS.md b/AGENTS.md index 9e5a683..e2a1d6b 100644 --- a/AGENTS.md +++ b/AGENTS.md @@ -22,7 +22,6 @@ Keep this file focused on cross-agent repo conventions: - output locations and file naming expectations - provenance and verification requirements - handoff rules between the lead agent and subagents -- remote delegation conventions Do **not** restate per-agent prompt text here unless there is a repo-wide constraint that applies to all agents. @@ -33,6 +32,7 @@ Do **not** restate per-agent prompt text here unless there is a repo-wide constr - Session logs go in `notes/`. - Plan artifacts for long-running workflows go in `outputs/.plans/`. - Intermediate research artifacts are written to disk by subagents and read by the lead agent. They are not returned inline unless the user explicitly asks for them. +- Long-running workflows should treat the plan artifact as an externalized working memory, not a static outline. Keep task status and verification state there as the run evolves. ## File naming @@ -54,14 +54,14 @@ Never use generic names like `research.md`, `draft.md`, `brief.md`, or `summary. - Provenance sidecars should record source accounting and verification status. - Source verification and citation cleanup belong in the `verifier` stage, not in ad hoc edits after delivery. - Verification passes should happen before delivery when the workflow calls for them. +- If a workflow uses the words `verified`, `confirmed`, or `checked`, the underlying artifact should record what was actually checked and how. +- For quantitative or code-backed outputs, keep raw artifact paths, scripts, or logs that support the final claim. Do not rely on polished summaries alone. +- Never smooth over missing checks. Mark work as `blocked`, `unverified`, or `inferred` when that is the honest status. ## Delegation rules - The lead agent plans, delegates, synthesizes, and delivers. - Use subagents when the work is meaningfully decomposable; do not spawn them for trivial work. - Prefer file-based handoffs over dumping large intermediate results back into parent context. -- When delegating to remote machines, retrieve final artifacts back into the local workspace and save them locally. - -## Remote delegation - -Feynman can delegate tasks to remote cloud machines via the `computer-fleet` and `computer-acp` skills. Load those skills on demand for CLI usage, session management, ACP bridging, and file retrieval. +- The lead agent is responsible for reconciling task completion. Subagents may not silently skip assigned tasks; skipped or merged tasks must be recorded in the plan artifact. +- For critical claims, require at least one adversarial verification pass after synthesis. Fix fatal issues before delivery or surface them explicitly. diff --git a/README.md b/README.md index 2e84b61..7bfba32 100644 --- a/README.md +++ b/README.md @@ -69,7 +69,6 @@ Four bundled research agents, dispatched automatically or via subagent commands. - **[AlphaXiv](https://www.alphaxiv.org/)** — paper search, Q&A, code reading, persistent annotations - **Docker** — isolated container execution for safe experiments on your machine -- **[Agent Computer](https://agentcomputer.ai)** — secure cloud execution for long-running research and GPU workloads - **Web search** — Gemini or Perplexity, zero-config default via signed-in Chromium - **Session search** — optional indexed recall across prior research sessions - **Preview** — browser and PDF export of generated artifacts @@ -95,7 +94,7 @@ feynman search status # web search config ## How it works -Built on [Pi](https://github.com/badlogic/pi-mono) for the agent runtime, [alphaXiv](https://www.alphaxiv.org/) for paper search and analysis, [Docker](https://www.docker.com/) for isolated local execution, and [Agent Computer](https://agentcomputer.ai) for secure cloud workloads +Built on [Pi](https://github.com/badlogic/pi-mono) for the agent runtime, [alphaXiv](https://www.alphaxiv.org/) for paper search and analysis, and [Docker](https://www.docker.com/) for isolated local execution Every output is source-grounded — claims link to papers, docs, or repos with direct URLs diff --git a/extensions/research-tools/preview.ts b/extensions/research-tools/preview.ts index 72d8821..cb6d7f7 100644 --- a/extensions/research-tools/preview.ts +++ b/extensions/research-tools/preview.ts @@ -181,53 +181,3 @@ export async function pathExists(path: string): Promise { return false; } } - -export function buildProjectAgentsTemplate(): string { - return `# Feynman Project Guide - -This file is read automatically at startup. It is the durable project memory for Feynman. - -## Project Overview -- State the research question, target artifact, target venue, and key datasets or benchmarks here. - -## AI Research Context -- Problem statement: -- Core hypothesis: -- Closest prior work: -- Required baselines: -- Required ablations: -- Primary metrics: -- Datasets / benchmarks: - -## Ground Rules -- Do not modify raw data in \`Data/Raw/\` or equivalent raw-data folders. -- Read first, act second: inspect project structure and existing notes before making changes. -- Prefer durable artifacts in \`notes/\`, \`outputs/\`, \`experiments/\`, and \`papers/\`. -- Keep strong claims source-grounded. Include direct URLs in final writeups. - -## Current Status -- Replace this section with the latest project status, known issues, and next steps. - -## Session Logging -- Use \`/log\` at the end of meaningful sessions to write a durable session note into \`notes/session-logs/\`. - -## Review Readiness -- Known reviewer concerns: -- Missing experiments: -- Missing writing or framing work: -`; -} - -export function buildSessionLogsReadme(): string { - return `# Session Logs - -Use \`/log\` to write one durable note per meaningful Feynman session. - -Recommended contents: -- what was done -- strongest findings -- artifacts written -- unresolved questions -- next steps -`; -} diff --git a/extensions/research-tools/project-scaffold.ts b/extensions/research-tools/project-scaffold.ts new file mode 100644 index 0000000..1db2708 --- /dev/null +++ b/extensions/research-tools/project-scaffold.ts @@ -0,0 +1,64 @@ +export function buildProjectAgentsTemplate(): string { + return `# Feynman Project Guide + +This file is read automatically at startup. It is the durable project memory for Feynman. + +## Project Overview +- State the research question, target artifact, target venue, and key datasets or benchmarks here. + +## AI Research Context +- Problem statement: +- Core hypothesis: +- Closest prior work: +- Required baselines: +- Required ablations: +- Primary metrics: +- Datasets / benchmarks: + +## Ground Rules +- Do not modify raw data in \`Data/Raw/\` or equivalent raw-data folders. +- Read first, act second: inspect project structure and existing notes before making changes. +- Prefer durable artifacts in \`notes/\`, \`outputs/\`, \`experiments/\`, and \`papers/\`. +- Keep strong claims source-grounded. Include direct URLs in final writeups. + +## Current Status +- Replace this section with the latest project status, known issues, and next steps. + +## Task Ledger +- Track concrete tasks with IDs, owner, status, and output path. +- Mark tasks as \`todo\`, \`in_progress\`, \`done\`, \`blocked\`, or \`superseded\`. +- Do not silently merge or skip tasks; record the decision here. + +## Verification Gates +- List the checks that must pass before delivery. +- For each critical claim, figure, or metric, record how it will be verified and where the raw artifact lives. +- Do not use words like \`verified\`, \`confirmed\`, or \`reproduced\` unless the underlying check actually ran. + +## Honesty Contract +- Separate direct observations from inferences. +- If something is uncertain, say so explicitly. +- If a result looks cleaner than expected, assume it needs another check before it goes into the final artifact. + +## Session Logging +- Use \`/log\` at the end of meaningful sessions to write a durable session note into \`notes/session-logs/\`. + +## Review Readiness +- Known reviewer concerns: +- Missing experiments: +- Missing writing or framing work: +`; +} + +export function buildSessionLogsReadme(): string { + return `# Session Logs + +Use \`/log\` to write one durable note per meaningful Feynman session. + +Recommended contents: +- what was done +- strongest findings +- artifacts written +- unresolved questions +- next steps +`; +} diff --git a/extensions/research-tools/project.ts b/extensions/research-tools/project.ts index 8601946..7830359 100644 --- a/extensions/research-tools/project.ts +++ b/extensions/research-tools/project.ts @@ -5,7 +5,8 @@ import type { ExtensionAPI } from "@mariozechner/pi-coding-agent"; import { Type } from "@sinclair/typebox"; import { getExtensionCommandSpec } from "../../metadata/commands.mjs"; -import { renderHtmlPreview, renderPdfPreview, openWithDefaultApp, pathExists, buildProjectAgentsTemplate, buildSessionLogsReadme } from "./preview.js"; +import { renderHtmlPreview, renderPdfPreview, openWithDefaultApp, pathExists } from "./preview.js"; +import { buildProjectAgentsTemplate, buildSessionLogsReadme } from "./project-scaffold.js"; import { formatToolText } from "./shared.js"; import { searchSessionTranscripts } from "./session-search.js"; diff --git a/prompts/autoresearch.md b/prompts/autoresearch.md index cc3604e..c61431c 100644 --- a/prompts/autoresearch.md +++ b/prompts/autoresearch.md @@ -26,7 +26,6 @@ Ask the user where to run: - **New git branch** — create a branch so main stays clean - **Virtual environment** — create an isolated venv/conda env first - **Docker** — run experiment code inside an isolated Docker container -- **Cloud** — delegate to a remote Agent Computer machine via `/delegate` Do not proceed without a clear answer. diff --git a/prompts/deepresearch.md b/prompts/deepresearch.md index 8c6a372..ac85bff 100644 --- a/prompts/deepresearch.md +++ b/prompts/deepresearch.md @@ -34,6 +34,16 @@ Derive a short slug from the topic (lowercase, hyphens, no filler words, ≤5 wo - [ ] Contradictions identified and addressed - [ ] No single-source claims on critical findings +## Task Ledger +| ID | Owner | Task | Status | Output | +|---|---|---|---|---| +| T1 | lead / researcher | ... | todo | ... | + +## Verification Log +| Item | Method | Status | Evidence | +|---|---|---|---| +| Critical claim / computation / figure | source cross-read / rerun / direct fetch / code check | pending | path or URL | + ## Decision Log (Updated as the workflow progresses) ``` @@ -60,6 +70,7 @@ Launch parallel `researcher` subagents via `subagent`. Each gets a structured br - **Output format:** numbered sources, evidence table, inline source references - **Tool guidance:** which search tools to prioritize - **Task boundaries:** what NOT to cover (another researcher handles that) +- **Task IDs:** the specific ledger rows they own and must report back on Assign each researcher a clearly disjoint dimension — different source types, geographic scopes, time periods, or technical angles. Never duplicate coverage. @@ -75,6 +86,7 @@ Assign each researcher a clearly disjoint dimension — different source types, ``` Researchers write full outputs to files and pass references back — do not have them return full content into your context. +Researchers must not silently merge or skip assigned tasks. If something is impossible or redundant, mark the ledger row `blocked` or `superseded` with a note. ## 4. Evaluate and loop @@ -83,10 +95,11 @@ After researchers return, read their output files and critically assess: - Which answers rest on only one source? - Are there contradictions needing resolution? - Is any key angle missing entirely? +- Did every assigned ledger task actually get completed, blocked, or explicitly superseded? If gaps are significant, spawn another targeted batch of researchers. No fixed cap on rounds — iterate until evidence is sufficient or sources are exhausted. -Update the plan artifact (`outputs/.plans/.md`) decision log after each round. +Update the plan artifact (`outputs/.plans/.md`) task ledger, verification log, and decision log after each round. Most topics need 1-2 rounds. Stop when additional rounds would not materially change conclusions. @@ -111,6 +124,12 @@ Unresolved issues, disagreements between sources, gaps in evidence. When the research includes quantitative data (benchmarks, performance comparisons, trends), generate charts using `pi-charts`. Use Mermaid diagrams for architectures and processes. Every visual must have a caption and reference the underlying data. +Before finalizing the draft, do a claim sweep: +- map each critical claim, number, and figure to its supporting source or artifact in the verification log +- downgrade or remove anything that cannot be grounded +- label inferences as inferences +- if code or calculations were involved, record which checks were actually run and which remain unverified + Save this draft to `outputs/.drafts/-draft.md`. ## 6. Cite @@ -136,6 +155,7 @@ Spawn the `reviewer` agent against the cited draft. The reviewer checks for: ``` If the reviewer flags FATAL issues, fix them in the brief before delivering. MAJOR issues get noted in the Open Questions section. MINOR issues are accepted. +After fixes, run at least one more review-style verification pass if any FATAL issues were found. Do not assume one fix solved everything. ## 8. Deliver diff --git a/prompts/delegate.md b/prompts/delegate.md deleted file mode 100644 index e462804..0000000 --- a/prompts/delegate.md +++ /dev/null @@ -1,21 +0,0 @@ ---- -description: Delegate a research task to a remote Agent Computer machine for cloud execution. -args: -section: Internal ---- -Delegate the following task to a remote Agent Computer machine: $@ - -## Workflow - -1. **Check CLI** — Verify `computer` or `aicomputer` is installed and authenticated. If not, install with `npm install -g aicomputer` and run `computer login`. -2. **Pick a machine** — Run `computer ls --json` and choose an appropriate machine. If none are running, tell the user to create one with `computer create`. -3. **Pick an agent** — Run `computer agent agents --json` and choose an installed agent with credentials (prefer Claude). -4. **Create a session** — Use `computer agent sessions new --agent claude --name research --json`. -5. **Send the task** — Translate the user's research task into a self-contained prompt and send it via `computer agent prompt`. The prompt must include: - - The full research objective - - Where to write outputs (default: `/workspace/outputs/`) - - What artifact to produce when done (summary file) - - Any tools or data sources to use -6. **Monitor** — Use `computer agent watch --session ` to stream progress. Report status to the user at meaningful milestones. -7. **Retrieve results** — When the remote agent finishes, pull the results back with `computer agent prompt "cat /workspace/outputs/.md" --session ` (derive the slug from the task topic). Present results to the user. -8. **Clean up** — Close the session with `computer agent close --session ` unless the user wants to continue. diff --git a/prompts/draft.md b/prompts/draft.md index 3a9a733..328b3cd 100644 --- a/prompts/draft.md +++ b/prompts/draft.md @@ -9,10 +9,11 @@ Write a paper-style draft for: $@ Derive a short slug from the topic (lowercase, hyphens, no filler words, ≤5 words). Use this slug for all files in this run. Requirements: -- Before writing, outline the draft structure: proposed title, sections, key claims to make, and source material to draw from. Write the outline to `outputs/.plans/.md`. Present the outline to the user and confirm before proceeding. +- Before writing, outline the draft structure: proposed title, sections, key claims to make, source material to draw from, and a verification log for the critical claims, figures, and calculations. Write the outline to `outputs/.plans/.md`. Present the outline to the user and confirm before proceeding. - Use the `writer` subagent when the draft should be produced from already-collected notes, then use the `verifier` subagent to add inline citations and verify sources. - Include at minimum: title, abstract, problem statement, related work, method or synthesis, evidence or experiments, limitations, conclusion. - Use clean Markdown with LaTeX where equations materially help. - Generate charts with `pi-charts` for quantitative data, benchmarks, and comparisons. Use Mermaid for architectures and pipelines. Every figure needs a caption. +- Before delivery, sweep the draft for any claim that sounds stronger than its support. Mark tentative results as tentative and remove unsupported numerics instead of letting the verifier discover them later. - Save exactly one draft to `papers/.md`. - End with a `Sources` appendix with direct URLs for all primary references. diff --git a/prompts/lit.md b/prompts/lit.md index 178ad69..d56d58b 100644 --- a/prompts/lit.md +++ b/prompts/lit.md @@ -10,9 +10,9 @@ Derive a short slug from the topic (lowercase, hyphens, no filler words, ≤5 wo ## Workflow -1. **Plan** — Outline the scope: key questions, source types to search (papers, web, repos), time period, and expected sections. Write the plan to `outputs/.plans/.md`. Present the plan to the user and confirm before proceeding. -2. **Gather** — Use the `researcher` subagent when the sweep is wide enough to benefit from delegated paper triage before synthesis. For narrow topics, search directly. Researcher outputs go to `-research-*.md`. -3. **Synthesize** — Separate consensus, disagreements, and open questions. When useful, propose concrete next experiments or follow-up reading. Generate charts with `pi-charts` for quantitative comparisons across papers and Mermaid diagrams for taxonomies or method pipelines. +1. **Plan** — Outline the scope: key questions, source types to search (papers, web, repos), time period, expected sections, and a small task ledger plus verification log. Write the plan to `outputs/.plans/.md`. Present the plan to the user and confirm before proceeding. +2. **Gather** — Use the `researcher` subagent when the sweep is wide enough to benefit from delegated paper triage before synthesis. For narrow topics, search directly. Researcher outputs go to `-research-*.md`. Do not silently skip assigned questions; mark them `done`, `blocked`, or `superseded`. +3. **Synthesize** — Separate consensus, disagreements, and open questions. When useful, propose concrete next experiments or follow-up reading. Generate charts with `pi-charts` for quantitative comparisons across papers and Mermaid diagrams for taxonomies or method pipelines. Before finishing the draft, sweep every strong claim against the verification log and downgrade anything that is inferred or single-source critical. 4. **Cite** — Spawn the `verifier` agent to add inline citations and verify every source URL in the draft. -5. **Verify** — Spawn the `reviewer` agent to check the cited draft for unsupported claims, logical gaps, and single-source critical findings. Fix FATAL issues before delivering. Note MAJOR issues in Open Questions. +5. **Verify** — Spawn the `reviewer` agent to check the cited draft for unsupported claims, logical gaps, zombie sections, and single-source critical findings. Fix FATAL issues before delivering. Note MAJOR issues in Open Questions. If FATAL issues were found, run one more verification pass after the fixes. 6. **Deliver** — Save the final literature review to `outputs/.md`. Write a provenance record alongside it as `outputs/.provenance.md` listing: date, sources consulted vs. accepted vs. rejected, verification status, and intermediate research files used. diff --git a/prompts/replicate.md b/prompts/replicate.md index 946bedf..0639f59 100644 --- a/prompts/replicate.md +++ b/prompts/replicate.md @@ -9,14 +9,13 @@ Design a replication plan for: $@ ## Workflow 1. **Extract** — Use the `researcher` subagent to pull implementation details from the target paper and any linked code. -2. **Plan** — Determine what code, datasets, metrics, and environment are needed. Be explicit about what is verified, what is inferred, and what is still missing. +2. **Plan** — Determine what code, datasets, metrics, and environment are needed. Be explicit about what is verified, what is inferred, what is still missing, and which checks or test oracles will be used to decide whether the replication succeeded. 3. **Environment** — Before running anything, ask the user where to execute: - **Local** — run in the current working directory - **Virtual environment** — create an isolated venv/conda env first - **Docker** — run experiment code inside an isolated Docker container - - **Cloud** — delegate to a remote Agent Computer machine via `/delegate` - **Plan only** — produce the replication plan without executing -4. **Execute** — If the user chose an execution environment, implement and run the replication steps there. Save notes, scripts, and results to disk in a reproducible layout. +4. **Execute** — If the user chose an execution environment, implement and run the replication steps there. Save notes, scripts, raw outputs, and results to disk in a reproducible layout. Do not call the outcome replicated unless the planned checks actually passed. 5. **Report** — End with a `Sources` section containing paper and repository URLs. Do not install packages, run training, or execute experiments without confirming the execution environment first. diff --git a/prompts/review.md b/prompts/review.md index aa1b1ee..dfa28a8 100644 --- a/prompts/review.md +++ b/prompts/review.md @@ -9,9 +9,10 @@ Review this AI research artifact: $@ Derive a short slug from the artifact name (lowercase, hyphens, no filler words, ≤5 words). Use this slug for all files in this run. Requirements: -- Before starting, outline what will be reviewed and the review criteria (novelty, empirical rigor, baselines, reproducibility, etc.). Present the plan to the user and confirm before proceeding. +- Before starting, outline what will be reviewed, the review criteria (novelty, empirical rigor, baselines, reproducibility, etc.), and any verification-specific checks needed for claims, figures, and reported metrics. Present the plan to the user and confirm before proceeding. - Spawn a `researcher` subagent to gather evidence on the artifact — inspect the paper, code, cited work, and any linked experimental artifacts. Save to `-research.md`. - Spawn a `reviewer` subagent with `-research.md` to produce the final peer review with inline annotations. - For small or simple artifacts where evidence gathering is overkill, run the `reviewer` subagent directly instead. +- If the first review finds FATAL issues and you fix them, run one more verification-style review pass before delivering. - Save exactly one review artifact to `outputs/-review.md`. - End with a `Sources` section containing direct URLs for every inspected external source. diff --git a/skills/agentcomputer/SKILL.md b/skills/agentcomputer/SKILL.md deleted file mode 100644 index 72ba5e3..0000000 --- a/skills/agentcomputer/SKILL.md +++ /dev/null @@ -1,108 +0,0 @@ ---- -name: agentcomputer -description: Delegate research tasks to remote Agent Computer machines for cloud execution. Manages machine discovery, remote agent sessions, task delegation, progress monitoring, result retrieval, and ACP bridging via the aicomputer CLI. -allowed-tools: Bash(npm:*), Bash(npx aicomputer@latest:*), Bash(aicomputer:*), Bash(computer:*) ---- - -# Agent Computer - -Use Agent Computer to run Feynman research workflows on remote cloud machines when local compute is insufficient or when tasks should run unattended. - -## When to use - -- A research task needs GPU, large memory, or long-running compute -- `/autoresearch` or `/deepresearch` should run unattended in the cloud -- The user explicitly asks to delegate work to a remote machine -- An experiment loop would take hours and should not block the local session - -## Prerequisites - -The `aicomputer` CLI must be installed and authenticated: - -```bash -if command -v computer >/dev/null 2>&1; then - COMPUTER=computer -elif command -v aicomputer >/dev/null 2>&1; then - COMPUTER=aicomputer -else - npm install -g aicomputer - COMPUTER=computer -fi -$COMPUTER whoami || $COMPUTER login -``` - -## Fleet control - -### Discover machines and agents - -```bash -$COMPUTER ls --json -$COMPUTER agent agents --json -``` - -### Sessions - -Create, reuse, and manage named sessions on a machine: - -```bash -$COMPUTER agent sessions new --agent claude --name research --json -$COMPUTER agent sessions list --json -$COMPUTER agent status --session --json -``` - -### Prompting and monitoring - -```bash -$COMPUTER agent prompt "" --agent claude --name research -$COMPUTER agent watch --session -``` - -### Stopping and cleanup - -```bash -$COMPUTER agent cancel --session --json -$COMPUTER agent interrupt --session --json -$COMPUTER agent close --session -``` - -## Research delegation workflow - -1. Pick a machine: `$COMPUTER ls --json` -2. Create a session: `$COMPUTER agent sessions new --agent claude --name research --json` -3. Send a self-contained research prompt: - -```bash -$COMPUTER agent prompt \ - "Run a deep research workflow on . Write all outputs to /workspace/outputs/. When done, write a summary to /workspace/outputs/summary.md." \ - --agent claude --name research -``` - -4. Monitor: `$COMPUTER agent watch --session ` -5. Retrieve: `$COMPUTER agent prompt "cat /workspace/outputs/summary.md" --session ` -6. Clean up: `$COMPUTER agent close --session ` - -## ACP bridge - -Expose a remote machine agent as a local ACP-compatible stdio process: - -```bash -$COMPUTER acp serve --agent claude --name research -``` - -This lets local ACP clients (including Feynman's subagents) talk to a remote agent as if it were local. Keep the bridge process running; reconnect by restarting the command with the same session name. - -## Session naming - -Use short stable names that match the task: - -- `research` — general research delegation -- `experiment` — autoresearch loops -- `review` — verification passes -- `literature` — literature sweeps - -Reuse the same name when continuing the same line of work. - -## References - -- [CLI cheatsheet](references/cli-cheatsheet.md) — full command reference -- [ACP flow](references/acp-flow.md) — protocol details for the ACP bridge diff --git a/skills/agentcomputer/references/acp-flow.md b/skills/agentcomputer/references/acp-flow.md deleted file mode 100644 index da626a6..0000000 --- a/skills/agentcomputer/references/acp-flow.md +++ /dev/null @@ -1,23 +0,0 @@ -# ACP Flow - -The `computer acp serve` bridge makes a remote machine agent look like a local ACP server over stdio. - -## Basic shape - -1. The local client starts `computer acp serve --agent --name `. -2. The bridge handles ACP initialization on stdin/stdout. -3. The bridge maps ACP session operations onto Agent Computer session APIs. -4. Remote session updates are streamed back as ACP `session/update` notifications. - -## Good commands - -```bash -computer acp serve my-box --agent claude --name research -computer acp serve gpu-worker --agent claude --name experiment -``` - -## Recommended client behavior - -- Reuse a stable session name when reconnecting. -- Treat the bridge as the single local command for remote-agent interaction. -- Use the normal `computer agent ...` commands outside ACP when you need manual inspection or cleanup. diff --git a/skills/agentcomputer/references/cli-cheatsheet.md b/skills/agentcomputer/references/cli-cheatsheet.md deleted file mode 100644 index 60d596c..0000000 --- a/skills/agentcomputer/references/cli-cheatsheet.md +++ /dev/null @@ -1,68 +0,0 @@ -# CLI Cheatsheet - -## Authentication - -```bash -computer whoami -computer login -computer claude-login # install Claude credentials on a machine -computer codex-login # install Codex credentials on a machine -``` - -## Machine discovery - -```bash -computer ls --json -computer fleet status --json -``` - -## Agent discovery - -```bash -computer agent agents --json -``` - -## Sessions - -```bash -computer agent sessions list --json -computer agent sessions new --agent claude --name research --json -computer agent status --session --json -``` - -## Prompting - -```bash -computer agent prompt "run the experiment" --agent claude --name research -computer agent prompt "continue" --session -``` - -## Streaming and control - -```bash -computer agent watch --session -computer agent cancel --session --json -computer agent interrupt --session --json -computer agent close --session -``` - -## ACP bridge - -```bash -computer acp serve --agent claude --name research -``` - -## Machine lifecycle - -```bash -computer create my-box -computer open my-box -computer open my-box --terminal -computer ssh my-box -``` - -## Good defaults - -- Prefer machine handles over machine ids when both are available. -- Prefer `--name` for human-meaningful persistent sessions. -- Prefer `--json` when another program or agent needs to read the result. diff --git a/website/.astro/data-store.json b/website/.astro/data-store.json deleted file mode 100644 index 287ae44..0000000 --- a/website/.astro/data-store.json +++ /dev/null @@ -1 +0,0 @@ -[["Map",1,2,9,10],"meta::meta",["Map",3,4,5,6,7,8],"astro-version","5.18.1","content-config-digest","d2da5d7c4a062d75","astro-config-digest","{\"root\":{},\"srcDir\":{},\"publicDir\":{},\"outDir\":{},\"cacheDir\":{},\"site\":\"https://feynman.companion.ai\",\"compressHTML\":true,\"base\":\"/\",\"trailingSlash\":\"ignore\",\"output\":\"static\",\"scopedStyleStrategy\":\"attribute\",\"build\":{\"format\":\"directory\",\"client\":{},\"server\":{},\"assets\":\"_astro\",\"serverEntry\":\"entry.mjs\",\"redirects\":true,\"inlineStylesheets\":\"auto\",\"concurrency\":1},\"server\":{\"open\":false,\"host\":false,\"port\":3001,\"streaming\":true,\"allowedHosts\":[]},\"redirects\":{},\"image\":{\"endpoint\":{\"route\":\"/_image\"},\"service\":{\"entrypoint\":\"astro/assets/services/sharp\",\"config\":{}},\"domains\":[],\"remotePatterns\":[],\"responsiveStyles\":false},\"devToolbar\":{\"enabled\":true},\"markdown\":{\"syntaxHighlight\":{\"type\":\"shiki\",\"excludeLangs\":[\"math\"]},\"shikiConfig\":{\"langs\":[],\"langAlias\":{},\"theme\":\"github-dark\",\"themes\":{\"light\":\"github-light\",\"dark\":\"github-dark\"},\"wrap\":false,\"transformers\":[]},\"remarkPlugins\":[],\"rehypePlugins\":[],\"remarkRehype\":{},\"gfm\":true,\"smartypants\":true},\"security\":{\"checkOrigin\":true,\"allowedDomains\":[],\"actionBodySizeLimit\":1048576},\"env\":{\"schema\":{},\"validateSecrets\":false},\"experimental\":{\"clientPrerender\":false,\"contentIntellisense\":false,\"headingIdCompat\":false,\"preserveScriptOrder\":false,\"liveContentCollections\":false,\"csp\":false,\"staticImportMetaEnv\":false,\"chromeDevtoolsWorkspace\":false,\"failOnPrerenderConflict\":false,\"svgo\":false},\"legacy\":{\"collections\":false}}","docs",["Map",11,12,58,59,89,90,132,133,171,172,200,201,236,237,269,270,298,299,319,320,345,346,375,376,398,399,423,424,452,453,484,485,510,511,532,533,553,554,577,578,599,600,623,624,644,645,668,669],"agents/reviewer",{"id":11,"data":13,"body":18,"filePath":19,"digest":20,"rendered":21,"legacyId":57},{"title":14,"description":15,"section":16,"order":17},"Reviewer","Simulate a tough but constructive AI research peer reviewer with inline annotations.","Agents",2,"## Source\n\nGenerated from `.feynman/agents/reviewer.md`. Edit that prompt file, not this docs page.\n\n## Role\n\nSimulate a tough but constructive AI research peer reviewer with inline annotations.\n\n## Default Output\n\n`review.md`\n\nYour job is to act like a skeptical but fair peer reviewer for AI/ML systems work.\n\n## Review checklist\n- Evaluate novelty, clarity, empirical rigor, reproducibility, and likely reviewer pushback.\n- Do not praise vaguely. Every positive claim should be tied to specific evidence.\n- Look for:\n - missing or weak baselines\n - missing ablations\n - evaluation mismatches\n - unclear claims of novelty\n - weak related-work positioning\n - insufficient statistical evidence\n - benchmark leakage or contamination risks\n - under-specified implementation details\n - claims that outrun the experiments\n- Distinguish between fatal issues, strong concerns, and polish issues.\n- Preserve uncertainty. If the draft might pass depending on venue norms, say so explicitly.\n\n## Output format\n\nProduce two sections: a structured review and inline annotations.\n\n### Part 1: Structured Review\n\n```markdown\n## Summary\n1-2 paragraph summary of the paper's contributions and approach.\n\n## Strengths\n- [S1] ...\n- [S2] ...\n\n## Weaknesses\n- [W1] **FATAL:** ...\n- [W2] **MAJOR:** ...\n- [W3] **MINOR:** ...\n\n## Questions for Authors\n- [Q1] ...\n\n## Verdict\nOverall assessment and confidence score. Would this pass at [venue]?\n\n## Revision Plan\nPrioritized, concrete steps to address each weakness.\n```\n\n### Part 2: Inline Annotations\n\nQuote specific passages from the paper and annotate them directly:\n\n```markdown\n## Inline Annotations\n\n> \"We achieve state-of-the-art results on all benchmarks\"\n**[W1] FATAL:** This claim is unsupported — Table 3 shows the method underperforms on 2 of 5 benchmarks. Revise to accurately reflect results.\n\n> \"Our approach is novel in combining X with Y\"\n**[W3] MINOR:** Z et al. (2024) combined X with Y in a different domain. Acknowledge this and clarify the distinction.\n\n> \"We use a learning rate of 1e-4\"\n**[Q1]:** Was this tuned? What range was searched? This matters for reproducibility.\n```\n\nReference the weakness/question IDs from Part 1 so annotations link back to the structured review.\n\n## Operating rules\n- Every weakness must reference a specific passage or section in the paper.\n- Inline annotations must quote the exact text being critiqued.\n- End with a `Sources` section containing direct URLs for anything additionally inspected during review.\n\n## Output contract\n- Save the main artifact to `review.md`.\n- The review must contain both the structured review AND inline annotations.","src/content/docs/agents/reviewer.md","115fe4b081dd8349",{"html":22,"metadata":23},"\u003Ch2 id=\"source\">Source\u003C/h2>\n\u003Cp>Generated from \u003Ccode>.feynman/agents/reviewer.md\u003C/code>. Edit that prompt file, not this docs page.\u003C/p>\n\u003Ch2 id=\"role\">Role\u003C/h2>\n\u003Cp>Simulate a tough but constructive AI research peer reviewer with inline annotations.\u003C/p>\n\u003Ch2 id=\"default-output\">Default Output\u003C/h2>\n\u003Cp>\u003Ccode>review.md\u003C/code>\u003C/p>\n\u003Cp>Your job is to act like a skeptical but fair peer reviewer for AI/ML systems work.\u003C/p>\n\u003Ch2 id=\"review-checklist\">Review checklist\u003C/h2>\n\u003Cul>\n\u003Cli>Evaluate novelty, clarity, empirical rigor, reproducibility, and likely reviewer pushback.\u003C/li>\n\u003Cli>Do not praise vaguely. Every positive claim should be tied to specific evidence.\u003C/li>\n\u003Cli>Look for:\n\u003Cul>\n\u003Cli>missing or weak baselines\u003C/li>\n\u003Cli>missing ablations\u003C/li>\n\u003Cli>evaluation mismatches\u003C/li>\n\u003Cli>unclear claims of novelty\u003C/li>\n\u003Cli>weak related-work positioning\u003C/li>\n\u003Cli>insufficient statistical evidence\u003C/li>\n\u003Cli>benchmark leakage or contamination risks\u003C/li>\n\u003Cli>under-specified implementation details\u003C/li>\n\u003Cli>claims that outrun the experiments\u003C/li>\n\u003C/ul>\n\u003C/li>\n\u003Cli>Distinguish between fatal issues, strong concerns, and polish issues.\u003C/li>\n\u003Cli>Preserve uncertainty. If the draft might pass depending on venue norms, say so explicitly.\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"output-format\">Output format\u003C/h2>\n\u003Cp>Produce two sections: a structured review and inline annotations.\u003C/p>\n\u003Ch3 id=\"part-1-structured-review\">Part 1: Structured Review\u003C/h3>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"markdown\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Summary\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">1-2 paragraph summary of the paper's contributions and approach.\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Strengths\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#E36209;--shiki-dark:#FFAB70\">-\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> [\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">S1\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">] ...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#E36209;--shiki-dark:#FFAB70\">-\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> [\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">S2\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">] ...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Weaknesses\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#E36209;--shiki-dark:#FFAB70\">-\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> [\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">W1\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">] \u003C/span>\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">**FATAL:**\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> ...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#E36209;--shiki-dark:#FFAB70\">-\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> [\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">W2\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">] \u003C/span>\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">**MAJOR:**\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> ...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#E36209;--shiki-dark:#FFAB70\">-\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> [\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">W3\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">] \u003C/span>\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">**MINOR:**\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> ...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Questions for Authors\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#E36209;--shiki-dark:#FFAB70\">-\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> [\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">Q1\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">] ...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Verdict\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">Overall assessment and confidence score. Would this pass at [\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">venue\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">]?\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Revision Plan\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">Prioritized, concrete steps to address each weakness.\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch3 id=\"part-2-inline-annotations\">Part 2: Inline Annotations\u003C/h3>\n\u003Cp>Quote specific passages from the paper and annotate them directly:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"markdown\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Inline Annotations\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#22863A;--shiki-dark:#85E89D\">> \"We achieve state-of-the-art results on all benchmarks\"\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">**[\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">W1\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">] FATAL:**\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> This claim is unsupported — Table 3 shows the method underperforms on 2 of 5 benchmarks. Revise to accurately reflect results.\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#22863A;--shiki-dark:#85E89D\">> \"Our approach is novel in combining X with Y\"\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">**[\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">W3\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">] MINOR:**\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> Z et al. (2024) combined X with Y in a different domain. Acknowledge this and clarify the distinction.\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#22863A;--shiki-dark:#85E89D\">> \"We use a learning rate of 1e-4\"\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">**[\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-light-text-decoration:underline;--shiki-dark:#DBEDFF;--shiki-dark-text-decoration:underline\">Q1\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-light-font-weight:bold;--shiki-dark:#E1E4E8;--shiki-dark-font-weight:bold\">]:**\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\"> Was this tuned? What range was searched? This matters for reproducibility.\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>Reference the weakness/question IDs from Part 1 so annotations link back to the structured review.\u003C/p>\n\u003Ch2 id=\"operating-rules\">Operating rules\u003C/h2>\n\u003Cul>\n\u003Cli>Every weakness must reference a specific passage or section in the paper.\u003C/li>\n\u003Cli>Inline annotations must quote the exact text being critiqued.\u003C/li>\n\u003Cli>End with a \u003Ccode>Sources\u003C/code> section containing direct URLs for anything additionally inspected during review.\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"output-contract\">Output contract\u003C/h2>\n\u003Cul>\n\u003Cli>Save the main artifact to \u003Ccode>review.md\u003C/code>.\u003C/li>\n\u003Cli>The review must contain both the structured review AND inline annotations.\u003C/li>\n\u003C/ul>",{"headings":24,"localImagePaths":53,"remoteImagePaths":54,"frontmatter":55,"imagePaths":56},[25,28,31,34,37,40,44,47,50],{"depth":17,"slug":26,"text":27},"source","Source",{"depth":17,"slug":29,"text":30},"role","Role",{"depth":17,"slug":32,"text":33},"default-output","Default Output",{"depth":17,"slug":35,"text":36},"review-checklist","Review checklist",{"depth":17,"slug":38,"text":39},"output-format","Output format",{"depth":41,"slug":42,"text":43},3,"part-1-structured-review","Part 1: Structured Review",{"depth":41,"slug":45,"text":46},"part-2-inline-annotations","Part 2: Inline Annotations",{"depth":17,"slug":48,"text":49},"operating-rules","Operating rules",{"depth":17,"slug":51,"text":52},"output-contract","Output contract",[],[],{"title":14,"description":15,"section":16,"order":17},[],"agents/reviewer.md","agents/writer",{"id":58,"data":60,"body":63,"filePath":64,"digest":65,"rendered":66,"legacyId":88},{"title":61,"description":62,"section":16,"order":41},"Writer","Turn research notes into clear, structured briefs and drafts.","## Source\n\nGenerated from `.feynman/agents/writer.md`. Edit that prompt file, not this docs page.\n\n## Role\n\nTurn research notes into clear, structured briefs and drafts.\n\n## Tools\n\n`read`, `bash`, `grep`, `find`, `ls`, `write`, `edit`\n\n## Default Output\n\n`draft.md`\n\n## Integrity commandments\n1. **Write only from supplied evidence.** Do not introduce claims, tools, or sources that are not in the input research files.\n2. **Preserve caveats and disagreements.** Never smooth away uncertainty.\n3. **Be explicit about gaps.** If the research files have unresolved questions or conflicting evidence, surface them — do not paper over them.\n\n## Output structure\n\n```markdown\n# Title\n\n## Executive Summary\n2-3 paragraph overview of key findings.\n\n## Section 1: ...\nDetailed findings organized by theme or question.\n\n## Section N: ...\n...\n\n## Open Questions\nUnresolved issues, disagreements between sources, gaps in evidence.\n```\n\n## Operating rules\n- Use clean Markdown structure and add equations only when they materially help.\n- Keep the narrative readable, but never outrun the evidence.\n- Produce artifacts that are ready to review in a browser or PDF preview.\n- Do NOT add inline citations — the verifier agent handles that as a separate post-processing step.\n- Do NOT add a Sources section — the verifier agent builds that.\n\n## Output contract\n- Save the main artifact to the specified output path (default: `draft.md`).\n- Focus on clarity, structure, and evidence traceability.","src/content/docs/agents/writer.md","ef9e81fb8113db70",{"html":67,"metadata":68},"\u003Ch2 id=\"source\">Source\u003C/h2>\n\u003Cp>Generated from \u003Ccode>.feynman/agents/writer.md\u003C/code>. Edit that prompt file, not this docs page.\u003C/p>\n\u003Ch2 id=\"role\">Role\u003C/h2>\n\u003Cp>Turn research notes into clear, structured briefs and drafts.\u003C/p>\n\u003Ch2 id=\"tools\">Tools\u003C/h2>\n\u003Cp>\u003Ccode>read\u003C/code>, \u003Ccode>bash\u003C/code>, \u003Ccode>grep\u003C/code>, \u003Ccode>find\u003C/code>, \u003Ccode>ls\u003C/code>, \u003Ccode>write\u003C/code>, \u003Ccode>edit\u003C/code>\u003C/p>\n\u003Ch2 id=\"default-output\">Default Output\u003C/h2>\n\u003Cp>\u003Ccode>draft.md\u003C/code>\u003C/p>\n\u003Ch2 id=\"integrity-commandments\">Integrity commandments\u003C/h2>\n\u003Col>\n\u003Cli>\u003Cstrong>Write only from supplied evidence.\u003C/strong> Do not introduce claims, tools, or sources that are not in the input research files.\u003C/li>\n\u003Cli>\u003Cstrong>Preserve caveats and disagreements.\u003C/strong> Never smooth away uncertainty.\u003C/li>\n\u003Cli>\u003Cstrong>Be explicit about gaps.\u003C/strong> If the research files have unresolved questions or conflicting evidence, surface them — do not paper over them.\u003C/li>\n\u003C/ol>\n\u003Ch2 id=\"output-structure\">Output structure\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"markdown\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\"># Title\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Executive Summary\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">2-3 paragraph overview of key findings.\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Section 1: ...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">Detailed findings organized by theme or question.\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Section N: ...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">...\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-light-font-weight:bold;--shiki-dark:#79B8FF;--shiki-dark-font-weight:bold\">## Open Questions\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">Unresolved issues, disagreements between sources, gaps in evidence.\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"operating-rules\">Operating rules\u003C/h2>\n\u003Cul>\n\u003Cli>Use clean Markdown structure and add equations only when they materially help.\u003C/li>\n\u003Cli>Keep the narrative readable, but never outrun the evidence.\u003C/li>\n\u003Cli>Produce artifacts that are ready to review in a browser or PDF preview.\u003C/li>\n\u003Cli>Do NOT add inline citations — the verifier agent handles that as a separate post-processing step.\u003C/li>\n\u003Cli>Do NOT add a Sources section — the verifier agent builds that.\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"output-contract\">Output contract\u003C/h2>\n\u003Cul>\n\u003Cli>Save the main artifact to the specified output path (default: \u003Ccode>draft.md\u003C/code>).\u003C/li>\n\u003Cli>Focus on clarity, structure, and evidence traceability.\u003C/li>\n\u003C/ul>",{"headings":69,"localImagePaths":84,"remoteImagePaths":85,"frontmatter":86,"imagePaths":87},[70,71,72,75,76,79,82,83],{"depth":17,"slug":26,"text":27},{"depth":17,"slug":29,"text":30},{"depth":17,"slug":73,"text":74},"tools","Tools",{"depth":17,"slug":32,"text":33},{"depth":17,"slug":77,"text":78},"integrity-commandments","Integrity commandments",{"depth":17,"slug":80,"text":81},"output-structure","Output structure",{"depth":17,"slug":48,"text":49},{"depth":17,"slug":51,"text":52},[],[],{"title":61,"description":62,"section":16,"order":41},[],"agents/writer.md","agents/researcher",{"id":89,"data":91,"body":95,"filePath":96,"digest":97,"rendered":98,"legacyId":131},{"title":92,"description":93,"section":16,"order":94},"Researcher","Gather primary evidence across papers, web sources, repos, docs, and local artifacts.",1,"## Source\n\nGenerated from `.feynman/agents/researcher.md`. Edit that prompt file, not this docs page.\n\n## Role\n\nGather primary evidence across papers, web sources, repos, docs, and local artifacts.\n\n## Tools\n\n`read`, `bash`, `grep`, `find`, `ls`\n\n## Default Output\n\n`research.md`\n\n## Integrity commandments\n1. **Never fabricate a source.** Every named tool, project, paper, product, or dataset must have a verifiable URL. If you cannot find a URL, do not mention it.\n2. **Never claim a project exists without checking.** Before citing a GitHub repo, search for it. Before citing a paper, find it. If a search returns zero results, the thing does not exist — do not invent it.\n3. **Never extrapolate details you haven't read.** If you haven't fetched and inspected a source, you may note its existence but must not describe its contents, metrics, or claims.\n4. **URL or it didn't happen.** Every entry in your evidence table must include a direct, checkable URL. No URL = not included.\n\n## Search strategy\n1. **Start wide.** Begin with short, broad queries to map the landscape. Use the `queries` array in `web_search` with 2–4 varied-angle queries simultaneously — never one query at a time when exploring.\n2. **Evaluate availability.** After the first round, assess what source types exist and which are highest quality. Adjust strategy accordingly.\n3. **Progressively narrow.** Drill into specifics using terminology and names discovered in initial results. Refine queries, don't repeat them.\n4. **Cross-source.** When the topic spans current reality and academic literature, always use both `web_search` and `alpha_search`.\n\nUse `recencyFilter` on `web_search` for fast-moving topics. Use `includeContent: true` on the most important results to get full page content rather than snippets.\n\n## Source quality\n- **Prefer:** academic papers, official documentation, primary datasets, verified benchmarks, government filings, reputable journalism, expert technical blogs, official vendor pages\n- **Accept with caveats:** well-cited secondary sources, established trade publications\n- **Deprioritize:** SEO-optimized listicles, undated blog posts, content aggregators, social media without primary links\n- **Reject:** sources with no author and no date, content that appears AI-generated with no primary backing\n\nWhen initial results skew toward low-quality sources, re-search with `domainFilter` targeting authoritative domains.\n\n## Output format\n\nAssign each source a stable numeric ID. Use these IDs consistently so downstream agents can trace claims to exact sources.\n\n### Evidence table\n\n| # | Source | URL | Key claim | Type | Confidence |\n|---|--------|-----|-----------|------|------------|\n| 1 | ... | ... | ... | primary / secondary / self-reported | high / medium / low |\n\n### Findings\n\nWrite findings using inline source references: `[1]`, `[2]`, etc. Every factual claim must cite at least one source by number.\n\n### Sources\n\nNumbered list matching the evidence table:\n1. Author/Title — URL\n2. Author/Title — URL\n\n## Context hygiene\n- Write findings to the output file progressively. Do not accumulate full page contents in your working memory — extract what you need, write it to file, move on.\n- When `includeContent: true` returns large pages, extract relevant quotes and discard the rest immediately.\n- If your search produces 10+ results, triage by title/snippet first. Only fetch full content for the top candidates.\n- Return a one-line summary to the parent, not full findings. The parent reads the output file.\n\n## Output contract\n- Save to the output file (default: `research.md`).\n- Minimum viable output: evidence table with ≥5 numbered entries, findings with inline references, and a numbered Sources section.\n- Write to the file and pass a lightweight reference back — do not dump full content into the parent context.","src/content/docs/agents/researcher.md","4d4d0e1b0fa38cd0",{"html":99,"metadata":100},"\u003Ch2 id=\"source\">Source\u003C/h2>\n\u003Cp>Generated from \u003Ccode>.feynman/agents/researcher.md\u003C/code>. Edit that prompt file, not this docs page.\u003C/p>\n\u003Ch2 id=\"role\">Role\u003C/h2>\n\u003Cp>Gather primary evidence across papers, web sources, repos, docs, and local artifacts.\u003C/p>\n\u003Ch2 id=\"tools\">Tools\u003C/h2>\n\u003Cp>\u003Ccode>read\u003C/code>, \u003Ccode>bash\u003C/code>, \u003Ccode>grep\u003C/code>, \u003Ccode>find\u003C/code>, \u003Ccode>ls\u003C/code>\u003C/p>\n\u003Ch2 id=\"default-output\">Default Output\u003C/h2>\n\u003Cp>\u003Ccode>research.md\u003C/code>\u003C/p>\n\u003Ch2 id=\"integrity-commandments\">Integrity commandments\u003C/h2>\n\u003Col>\n\u003Cli>\u003Cstrong>Never fabricate a source.\u003C/strong> Every named tool, project, paper, product, or dataset must have a verifiable URL. If you cannot find a URL, do not mention it.\u003C/li>\n\u003Cli>\u003Cstrong>Never claim a project exists without checking.\u003C/strong> Before citing a GitHub repo, search for it. Before citing a paper, find it. If a search returns zero results, the thing does not exist — do not invent it.\u003C/li>\n\u003Cli>\u003Cstrong>Never extrapolate details you haven’t read.\u003C/strong> If you haven’t fetched and inspected a source, you may note its existence but must not describe its contents, metrics, or claims.\u003C/li>\n\u003Cli>\u003Cstrong>URL or it didn’t happen.\u003C/strong> Every entry in your evidence table must include a direct, checkable URL. No URL = not included.\u003C/li>\n\u003C/ol>\n\u003Ch2 id=\"search-strategy\">Search strategy\u003C/h2>\n\u003Col>\n\u003Cli>\u003Cstrong>Start wide.\u003C/strong> Begin with short, broad queries to map the landscape. Use the \u003Ccode>queries\u003C/code> array in \u003Ccode>web_search\u003C/code> with 2–4 varied-angle queries simultaneously — never one query at a time when exploring.\u003C/li>\n\u003Cli>\u003Cstrong>Evaluate availability.\u003C/strong> After the first round, assess what source types exist and which are highest quality. Adjust strategy accordingly.\u003C/li>\n\u003Cli>\u003Cstrong>Progressively narrow.\u003C/strong> Drill into specifics using terminology and names discovered in initial results. Refine queries, don’t repeat them.\u003C/li>\n\u003Cli>\u003Cstrong>Cross-source.\u003C/strong> When the topic spans current reality and academic literature, always use both \u003Ccode>web_search\u003C/code> and \u003Ccode>alpha_search\u003C/code>.\u003C/li>\n\u003C/ol>\n\u003Cp>Use \u003Ccode>recencyFilter\u003C/code> on \u003Ccode>web_search\u003C/code> for fast-moving topics. Use \u003Ccode>includeContent: true\u003C/code> on the most important results to get full page content rather than snippets.\u003C/p>\n\u003Ch2 id=\"source-quality\">Source quality\u003C/h2>\n\u003Cul>\n\u003Cli>\u003Cstrong>Prefer:\u003C/strong> academic papers, official documentation, primary datasets, verified benchmarks, government filings, reputable journalism, expert technical blogs, official vendor pages\u003C/li>\n\u003Cli>\u003Cstrong>Accept with caveats:\u003C/strong> well-cited secondary sources, established trade publications\u003C/li>\n\u003Cli>\u003Cstrong>Deprioritize:\u003C/strong> SEO-optimized listicles, undated blog posts, content aggregators, social media without primary links\u003C/li>\n\u003Cli>\u003Cstrong>Reject:\u003C/strong> sources with no author and no date, content that appears AI-generated with no primary backing\u003C/li>\n\u003C/ul>\n\u003Cp>When initial results skew toward low-quality sources, re-search with \u003Ccode>domainFilter\u003C/code> targeting authoritative domains.\u003C/p>\n\u003Ch2 id=\"output-format\">Output format\u003C/h2>\n\u003Cp>Assign each source a stable numeric ID. Use these IDs consistently so downstream agents can trace claims to exact sources.\u003C/p>\n\u003Ch3 id=\"evidence-table\">Evidence table\u003C/h3>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>#\u003C/th>\u003Cth>Source\u003C/th>\u003Cth>URL\u003C/th>\u003Cth>Key claim\u003C/th>\u003Cth>Type\u003C/th>\u003Cth>Confidence\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>1\u003C/td>\u003Ctd>…\u003C/td>\u003Ctd>…\u003C/td>\u003Ctd>…\u003C/td>\u003Ctd>primary / secondary / self-reported\u003C/td>\u003Ctd>high / medium / low\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch3 id=\"findings\">Findings\u003C/h3>\n\u003Cp>Write findings using inline source references: \u003Ccode>[1]\u003C/code>, \u003Ccode>[2]\u003C/code>, etc. Every factual claim must cite at least one source by number.\u003C/p>\n\u003Ch3 id=\"sources\">Sources\u003C/h3>\n\u003Cp>Numbered list matching the evidence table:\u003C/p>\n\u003Col>\n\u003Cli>Author/Title — URL\u003C/li>\n\u003Cli>Author/Title — URL\u003C/li>\n\u003C/ol>\n\u003Ch2 id=\"context-hygiene\">Context hygiene\u003C/h2>\n\u003Cul>\n\u003Cli>Write findings to the output file progressively. Do not accumulate full page contents in your working memory — extract what you need, write it to file, move on.\u003C/li>\n\u003Cli>When \u003Ccode>includeContent: true\u003C/code> returns large pages, extract relevant quotes and discard the rest immediately.\u003C/li>\n\u003Cli>If your search produces 10+ results, triage by title/snippet first. Only fetch full content for the top candidates.\u003C/li>\n\u003Cli>Return a one-line summary to the parent, not full findings. The parent reads the output file.\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"output-contract\">Output contract\u003C/h2>\n\u003Cul>\n\u003Cli>Save to the output file (default: \u003Ccode>research.md\u003C/code>).\u003C/li>\n\u003Cli>Minimum viable output: evidence table with ≥5 numbered entries, findings with inline references, and a numbered Sources section.\u003C/li>\n\u003Cli>Write to the file and pass a lightweight reference back — do not dump full content into the parent context.\u003C/li>\n\u003C/ul>",{"headings":101,"localImagePaths":127,"remoteImagePaths":128,"frontmatter":129,"imagePaths":130},[102,103,104,105,106,107,110,113,114,117,120,123,126],{"depth":17,"slug":26,"text":27},{"depth":17,"slug":29,"text":30},{"depth":17,"slug":73,"text":74},{"depth":17,"slug":32,"text":33},{"depth":17,"slug":77,"text":78},{"depth":17,"slug":108,"text":109},"search-strategy","Search strategy",{"depth":17,"slug":111,"text":112},"source-quality","Source quality",{"depth":17,"slug":38,"text":39},{"depth":41,"slug":115,"text":116},"evidence-table","Evidence table",{"depth":41,"slug":118,"text":119},"findings","Findings",{"depth":41,"slug":121,"text":122},"sources","Sources",{"depth":17,"slug":124,"text":125},"context-hygiene","Context hygiene",{"depth":17,"slug":51,"text":52},[],[],{"title":92,"description":93,"section":16,"order":94},[],"agents/researcher.md","getting-started/setup",{"id":132,"data":134,"body":138,"filePath":139,"digest":140,"rendered":141,"legacyId":170},{"title":135,"description":136,"section":137,"order":41},"Setup","Detailed setup guide for Feynman","Getting Started","## Guided setup\n\n```bash\nfeynman setup\n```\n\nThis walks through four steps:\n\n### Model provider authentication\n\nFeynman uses Pi's OAuth system for model access. The setup wizard prompts you to log in to your preferred provider.\n\n```bash\nfeynman model login\n```\n\n### AlphaXiv login\n\nAlphaXiv powers Feynman's paper search and analysis tools. Sign in with:\n\n```bash\nfeynman alpha login\n```\n\nCheck status anytime:\n\n```bash\nfeynman alpha status\n```\n\n### Web search routing\n\nFeynman supports three web search backends:\n\n- **auto** — Prefer Perplexity when configured, fall back to Gemini\n- **perplexity** — Force Perplexity Sonar\n- **gemini** — Force Gemini (default, zero-config via signed-in Chromium)\n\nThe default path requires no API keys — it uses Gemini Browser via your signed-in Chromium profile.\n\n### Preview dependencies\n\nFor PDF and HTML export of generated artifacts, Feynman needs `pandoc`:\n\n```bash\nfeynman --setup-preview\n```\n\nGlobal macOS installs also try to install pandoc automatically when Homebrew is available. Use the command above to retry manually.\n\n### Optional packages\n\nFeynman keeps the default package set lean so first-run installs stay fast. Install the heavier optional packages only when you need them:\n\n```bash\nfeynman packages list\nfeynman packages install memory\nfeynman packages install session-search\nfeynman packages install generative-ui\nfeynman packages install all-extras\n```\n\n## Diagnostics\n\nRun the doctor to check everything:\n\n```bash\nfeynman doctor\n```\n\nThis verifies model auth, alphaXiv credentials, preview dependencies, and the Pi runtime.","src/content/docs/getting-started/setup.md","ac3f3bae92f2beeb",{"html":142,"metadata":143},"\u003Ch2 id=\"guided-setup\">Guided setup\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> setup\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>This walks through four steps:\u003C/p>\n\u003Ch3 id=\"model-provider-authentication\">Model provider authentication\u003C/h3>\n\u003Cp>Feynman uses Pi’s OAuth system for model access. The setup wizard prompts you to log in to your preferred provider.\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> model\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> login\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch3 id=\"alphaxiv-login\">AlphaXiv login\u003C/h3>\n\u003Cp>AlphaXiv powers Feynman’s paper search and analysis tools. Sign in with:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> alpha\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> login\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>Check status anytime:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> alpha\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> status\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch3 id=\"web-search-routing\">Web search routing\u003C/h3>\n\u003Cp>Feynman supports three web search backends:\u003C/p>\n\u003Cul>\n\u003Cli>\u003Cstrong>auto\u003C/strong> — Prefer Perplexity when configured, fall back to Gemini\u003C/li>\n\u003Cli>\u003Cstrong>perplexity\u003C/strong> — Force Perplexity Sonar\u003C/li>\n\u003Cli>\u003Cstrong>gemini\u003C/strong> — Force Gemini (default, zero-config via signed-in Chromium)\u003C/li>\n\u003C/ul>\n\u003Cp>The default path requires no API keys — it uses Gemini Browser via your signed-in Chromium profile.\u003C/p>\n\u003Ch3 id=\"preview-dependencies\">Preview dependencies\u003C/h3>\n\u003Cp>For PDF and HTML export of generated artifacts, Feynman needs \u003Ccode>pandoc\u003C/code>:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> --setup-preview\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>Global macOS installs also try to install pandoc automatically when Homebrew is available. Use the command above to retry manually.\u003C/p>\n\u003Ch3 id=\"optional-packages\">Optional packages\u003C/h3>\n\u003Cp>Feynman keeps the default package set lean so first-run installs stay fast. Install the heavier optional packages only when you need them:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> packages\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> list\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> packages\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> install\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> memory\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> packages\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> install\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> session-search\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> packages\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> install\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> generative-ui\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> packages\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> install\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> all-extras\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"diagnostics\">Diagnostics\u003C/h2>\n\u003Cp>Run the doctor to check everything:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> doctor\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>This verifies model auth, alphaXiv credentials, preview dependencies, and the Pi runtime.\u003C/p>",{"headings":144,"localImagePaths":166,"remoteImagePaths":167,"frontmatter":168,"imagePaths":169},[145,148,151,154,157,160,163],{"depth":17,"slug":146,"text":147},"guided-setup","Guided setup",{"depth":41,"slug":149,"text":150},"model-provider-authentication","Model provider authentication",{"depth":41,"slug":152,"text":153},"alphaxiv-login","AlphaXiv login",{"depth":41,"slug":155,"text":156},"web-search-routing","Web search routing",{"depth":41,"slug":158,"text":159},"preview-dependencies","Preview dependencies",{"depth":41,"slug":161,"text":162},"optional-packages","Optional packages",{"depth":17,"slug":164,"text":165},"diagnostics","Diagnostics",[],[],{"title":135,"description":136,"section":137,"order":41},[],"getting-started/setup.md","getting-started/quickstart",{"id":171,"data":173,"body":176,"filePath":177,"digest":178,"rendered":179,"legacyId":199},{"title":174,"description":175,"section":137,"order":17},"Quick Start","Get up and running with Feynman in 60 seconds","## First run\n\n```bash\nfeynman setup\nfeynman\n```\n\n`feynman setup` walks you through model authentication, alphaXiv login, web search configuration, and preview dependencies.\n\n## Ask naturally\n\nFeynman routes your questions into the right workflow automatically. You don't need slash commands to get started.\n\n```\n> What are the main approaches to RLHF alignment?\n```\n\nFeynman will search papers, gather web sources, and produce a structured answer with citations.\n\n## Use workflows directly\n\nFor explicit control, use slash commands inside the REPL:\n\n```\n> /deepresearch transformer scaling laws\n> /lit multimodal reasoning benchmarks\n> /review paper.pdf\n```\n\n## Output locations\n\nFeynman writes durable artifacts to canonical directories:\n\n- `outputs/` — Reviews, reading lists, summaries\n- `papers/` — Polished paper-style drafts\n- `experiments/` — Runnable code and result logs\n- `notes/` — Scratch notes and session logs","src/content/docs/getting-started/quickstart.md","0a22caade9f6c5a5",{"html":180,"metadata":181},"\u003Ch2 id=\"first-run\">First run\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> setup\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>\u003Ccode>feynman setup\u003C/code> walks you through model authentication, alphaXiv login, web search configuration, and preview dependencies.\u003C/p>\n\u003Ch2 id=\"ask-naturally\">Ask naturally\u003C/h2>\n\u003Cp>Feynman routes your questions into the right workflow automatically. You don’t need slash commands to get started.\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>> What are the main approaches to RLHF alignment?\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>Feynman will search papers, gather web sources, and produce a structured answer with citations.\u003C/p>\n\u003Ch2 id=\"use-workflows-directly\">Use workflows directly\u003C/h2>\n\u003Cp>For explicit control, use slash commands inside the REPL:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>> /deepresearch transformer scaling laws\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan>> /lit multimodal reasoning benchmarks\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan>> /review paper.pdf\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"output-locations\">Output locations\u003C/h2>\n\u003Cp>Feynman writes durable artifacts to canonical directories:\u003C/p>\n\u003Cul>\n\u003Cli>\u003Ccode>outputs/\u003C/code> — Reviews, reading lists, summaries\u003C/li>\n\u003Cli>\u003Ccode>papers/\u003C/code> — Polished paper-style drafts\u003C/li>\n\u003Cli>\u003Ccode>experiments/\u003C/code> — Runnable code and result logs\u003C/li>\n\u003Cli>\u003Ccode>notes/\u003C/code> — Scratch notes and session logs\u003C/li>\n\u003C/ul>",{"headings":182,"localImagePaths":195,"remoteImagePaths":196,"frontmatter":197,"imagePaths":198},[183,186,189,192],{"depth":17,"slug":184,"text":185},"first-run","First run",{"depth":17,"slug":187,"text":188},"ask-naturally","Ask naturally",{"depth":17,"slug":190,"text":191},"use-workflows-directly","Use workflows directly",{"depth":17,"slug":193,"text":194},"output-locations","Output locations",[],[],{"title":174,"description":175,"section":137,"order":17},[],"getting-started/quickstart.md","getting-started/configuration",{"id":200,"data":202,"body":206,"filePath":207,"digest":208,"rendered":209,"legacyId":235},{"title":203,"description":204,"section":137,"order":205},"Configuration","Configure models, search, and runtime options",4,"## Model\n\nSet the default model:\n\n```bash\nfeynman model set \u003Cprovider:model>\n```\n\nOverride at runtime:\n\n```bash\nfeynman --model anthropic:claude-opus-4-6\n```\n\nList available models:\n\n```bash\nfeynman model list\n```\n\n## Thinking level\n\nControl the reasoning depth:\n\n```bash\nfeynman --thinking high\n```\n\nLevels: `off`, `minimal`, `low`, `medium`, `high`, `xhigh`.\n\n## Web search\n\nCheck the current search configuration:\n\n```bash\nfeynman search status\n```\n\nFor advanced configuration, edit `~/.feynman/web-search.json` directly to set Gemini API keys, Perplexity keys, or a different route.\n\n## Working directory\n\n```bash\nfeynman --cwd /path/to/project\n```\n\n## Session storage\n\n```bash\nfeynman --session-dir /path/to/sessions\n```\n\n## One-shot mode\n\nRun a single prompt and exit:\n\n```bash\nfeynman --prompt \"summarize the key findings of 2401.12345\"\n```","src/content/docs/getting-started/configuration.md","9d66eb82ad4b948a",{"html":210,"metadata":211},"\u003Ch2 id=\"model\">Model\u003C/h2>\n\u003Cp>Set the default model:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> model\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> set\u003C/span>\u003Cspan style=\"color:#D73A49;--shiki-dark:#F97583\"> <\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\">provider:mode\u003C/span>\u003Cspan style=\"color:#24292E;--shiki-dark:#E1E4E8\">l\u003C/span>\u003Cspan style=\"color:#D73A49;--shiki-dark:#F97583\">>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>Override at runtime:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> --model\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> anthropic:claude-opus-4-6\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>List available models:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> model\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> list\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"thinking-level\">Thinking level\u003C/h2>\n\u003Cp>Control the reasoning depth:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> --thinking\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> high\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>Levels: \u003Ccode>off\u003C/code>, \u003Ccode>minimal\u003C/code>, \u003Ccode>low\u003C/code>, \u003Ccode>medium\u003C/code>, \u003Ccode>high\u003C/code>, \u003Ccode>xhigh\u003C/code>.\u003C/p>\n\u003Ch2 id=\"web-search\">Web search\u003C/h2>\n\u003Cp>Check the current search configuration:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> search\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> status\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>For advanced configuration, edit \u003Ccode>~/.feynman/web-search.json\u003C/code> directly to set Gemini API keys, Perplexity keys, or a different route.\u003C/p>\n\u003Ch2 id=\"working-directory\">Working directory\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> --cwd\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> /path/to/project\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"session-storage\">Session storage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> --session-dir\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> /path/to/sessions\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"one-shot-mode\">One-shot mode\u003C/h2>\n\u003Cp>Run a single prompt and exit:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> --prompt\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> \"summarize the key findings of 2401.12345\"\u003C/span>\u003C/span>\u003C/code>\u003C/pre>",{"headings":212,"localImagePaths":231,"remoteImagePaths":232,"frontmatter":233,"imagePaths":234},[213,216,219,222,225,228],{"depth":17,"slug":214,"text":215},"model","Model",{"depth":17,"slug":217,"text":218},"thinking-level","Thinking level",{"depth":17,"slug":220,"text":221},"web-search","Web search",{"depth":17,"slug":223,"text":224},"working-directory","Working directory",{"depth":17,"slug":226,"text":227},"session-storage","Session storage",{"depth":17,"slug":229,"text":230},"one-shot-mode","One-shot mode",[],[],{"title":203,"description":204,"section":137,"order":205},[],"getting-started/configuration.md","reference/cli-commands",{"id":236,"data":238,"body":242,"filePath":243,"digest":244,"rendered":245,"legacyId":268},{"title":239,"description":240,"section":241,"order":94},"CLI Commands","Complete reference for Feynman CLI commands","Reference","This page covers the dedicated Feynman CLI commands and compatibility flags.\n\nWorkflow prompt templates such as `/deepresearch` also run directly from the shell as `feynman \u003Cworkflow> ...`. Those workflow entries live in the slash-command reference instead of being duplicated here.\n\n## Core\n\n| Command | Description |\n| --- | --- |\n| `feynman` | Launch the interactive REPL. |\n| `feynman chat [prompt]` | Start chat explicitly, optionally with an initial prompt. |\n| `feynman help` | Show CLI help. |\n| `feynman setup` | Run the guided setup wizard. |\n| `feynman doctor` | Diagnose config, auth, Pi runtime, and preview dependencies. |\n| `feynman status` | Show the current setup summary. |\n\n## Model Management\n\n| Command | Description |\n| --- | --- |\n| `feynman model list` | List available models in Pi auth storage. |\n| `feynman model login [id]` | Login to a Pi OAuth model provider. |\n| `feynman model logout [id]` | Logout from a Pi OAuth model provider. |\n| `feynman model set \u003Cprovider/model>` | Set the default model. |\n\n## AlphaXiv\n\n| Command | Description |\n| --- | --- |\n| `feynman alpha login` | Sign in to alphaXiv. |\n| `feynman alpha logout` | Clear alphaXiv auth. |\n| `feynman alpha status` | Check alphaXiv auth status. |\n\n## Utilities\n\n| Command | Description |\n| --- | --- |\n| `feynman search status` | Show Pi web-access status and config path. |\n| `feynman update [package]` | Update installed packages, or a specific package. |\n\n## Flags\n\n| Flag | Description |\n| --- | --- |\n| `--prompt \"\u003Ctext>\"` | Run one prompt and exit. |\n| `--alpha-login` | Sign in to alphaXiv and exit. |\n| `--alpha-logout` | Clear alphaXiv auth and exit. |\n| `--alpha-status` | Show alphaXiv auth status and exit. |\n| `--model \u003Cprovider:model>` | Force a specific model. |\n| `--thinking \u003Clevel>` | Set thinking level: off | minimal | low | medium | high | xhigh. |\n| `--cwd \u003Cpath>` | Set the working directory for tools. |\n| `--session-dir \u003Cpath>` | Set the session storage directory. |\n| `--new-session` | Start a new persisted session. |\n| `--doctor` | Alias for `feynman doctor`. |\n| `--setup-preview` | Alias for `feynman setup preview`. |","src/content/docs/reference/cli-commands.md","5ba10666ccf260a6",{"html":246,"metadata":247},"\u003Cp>This page covers the dedicated Feynman CLI commands and compatibility flags.\u003C/p>\n\u003Cp>Workflow prompt templates such as \u003Ccode>/deepresearch\u003C/code> also run directly from the shell as \u003Ccode>feynman <workflow> ...\u003C/code>. Those workflow entries live in the slash-command reference instead of being duplicated here.\u003C/p>\n\u003Ch2 id=\"core\">Core\u003C/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Command\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>feynman\u003C/code>\u003C/td>\u003Ctd>Launch the interactive REPL.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman chat [prompt]\u003C/code>\u003C/td>\u003Ctd>Start chat explicitly, optionally with an initial prompt.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman help\u003C/code>\u003C/td>\u003Ctd>Show CLI help.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman setup\u003C/code>\u003C/td>\u003Ctd>Run the guided setup wizard.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman doctor\u003C/code>\u003C/td>\u003Ctd>Diagnose config, auth, Pi runtime, and preview dependencies.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman status\u003C/code>\u003C/td>\u003Ctd>Show the current setup summary.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch2 id=\"model-management\">Model Management\u003C/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Command\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>feynman model list\u003C/code>\u003C/td>\u003Ctd>List available models in Pi auth storage.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman model login [id]\u003C/code>\u003C/td>\u003Ctd>Login to a Pi OAuth model provider.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman model logout [id]\u003C/code>\u003C/td>\u003Ctd>Logout from a Pi OAuth model provider.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman model set <provider/model>\u003C/code>\u003C/td>\u003Ctd>Set the default model.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch2 id=\"alphaxiv\">AlphaXiv\u003C/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Command\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>feynman alpha login\u003C/code>\u003C/td>\u003Ctd>Sign in to alphaXiv.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman alpha logout\u003C/code>\u003C/td>\u003Ctd>Clear alphaXiv auth.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman alpha status\u003C/code>\u003C/td>\u003Ctd>Check alphaXiv auth status.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch2 id=\"utilities\">Utilities\u003C/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Command\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>feynman search status\u003C/code>\u003C/td>\u003Ctd>Show Pi web-access status and config path.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>feynman update [package]\u003C/code>\u003C/td>\u003Ctd>Update installed packages, or a specific package.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch2 id=\"flags\">Flags\u003C/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Flag\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>--prompt \"<text>\"\u003C/code>\u003C/td>\u003Ctd>Run one prompt and exit.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--alpha-login\u003C/code>\u003C/td>\u003Ctd>Sign in to alphaXiv and exit.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--alpha-logout\u003C/code>\u003C/td>\u003Ctd>Clear alphaXiv auth and exit.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--alpha-status\u003C/code>\u003C/td>\u003Ctd>Show alphaXiv auth status and exit.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--model <provider:model>\u003C/code>\u003C/td>\u003Ctd>Force a specific model.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--thinking <level>\u003C/code>\u003C/td>\u003Ctd>Set thinking level: off\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--cwd <path>\u003C/code>\u003C/td>\u003Ctd>Set the working directory for tools.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--session-dir <path>\u003C/code>\u003C/td>\u003Ctd>Set the session storage directory.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--new-session\u003C/code>\u003C/td>\u003Ctd>Start a new persisted session.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--doctor\u003C/code>\u003C/td>\u003Ctd>Alias for \u003Ccode>feynman doctor\u003C/code>.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>--setup-preview\u003C/code>\u003C/td>\u003Ctd>Alias for \u003Ccode>feynman setup preview\u003C/code>.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>",{"headings":248,"localImagePaths":264,"remoteImagePaths":265,"frontmatter":266,"imagePaths":267},[249,252,255,258,261],{"depth":17,"slug":250,"text":251},"core","Core",{"depth":17,"slug":253,"text":254},"model-management","Model Management",{"depth":17,"slug":256,"text":257},"alphaxiv","AlphaXiv",{"depth":17,"slug":259,"text":260},"utilities","Utilities",{"depth":17,"slug":262,"text":263},"flags","Flags",[],[],{"title":239,"description":240,"section":241,"order":94},[],"reference/cli-commands.md","getting-started/installation",{"id":269,"data":271,"body":274,"filePath":275,"digest":276,"rendered":277,"legacyId":297},{"title":272,"description":273,"section":137,"order":94},"Installation","Install Feynman and get started","## Requirements\n\n- Node.js 20 or later\n- npm 9 or later\n\n## Install\n\n```bash\nnpm install -g @companion-ai/feynman\n```\n\n## Verify\n\n```bash\nfeynman --version\n```\n\n## Local Development\n\nFor contributing or local development:\n\n```bash\ngit clone https://github.com/getcompanion-ai/feynman.git\ncd feynman\nnpm install\nnpm run start\n```","src/content/docs/getting-started/installation.md","781ab0278b8c1673",{"html":278,"metadata":279},"\u003Ch2 id=\"requirements\">Requirements\u003C/h2>\n\u003Cul>\n\u003Cli>Node.js 20 or later\u003C/li>\n\u003Cli>npm 9 or later\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"install\">Install\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">npm\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> install\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> -g\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> @companion-ai/feynman\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"verify\">Verify\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> --version\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"local-development\">Local Development\u003C/h2>\n\u003Cp>For contributing or local development:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">git\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> clone\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> https://github.com/getcompanion-ai/feynman.git\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\">cd\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> feynman\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">npm\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> install\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">npm\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> run\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> start\u003C/span>\u003C/span>\u003C/code>\u003C/pre>",{"headings":280,"localImagePaths":293,"remoteImagePaths":294,"frontmatter":295,"imagePaths":296},[281,284,287,290],{"depth":17,"slug":282,"text":283},"requirements","Requirements",{"depth":17,"slug":285,"text":286},"install","Install",{"depth":17,"slug":288,"text":289},"verify","Verify",{"depth":17,"slug":291,"text":292},"local-development","Local Development",[],[],{"title":272,"description":273,"section":137,"order":94},[],"getting-started/installation.md","reference/package-stack",{"id":298,"data":300,"body":303,"filePath":304,"digest":305,"rendered":306,"legacyId":318},{"title":301,"description":302,"section":241,"order":41},"Package Stack","Curated Pi packages bundled with Feynman","Curated Pi packages bundled with Feynman. The runtime package list lives in `.feynman/settings.json`.\n\n## Core packages\n\nInstalled by default.\n\n| Package | Purpose |\n|---------|---------|\n| `pi-subagents` | Parallel literature gathering and decomposition. |\n| `pi-btw` | Fast side-thread `/btw` conversations without interrupting the main run. |\n| `pi-docparser` | PDFs, Office docs, spreadsheets, and images. |\n| `pi-web-access` | Web, GitHub, PDF, and media access. |\n| `pi-markdown-preview` | Polished Markdown and LaTeX-heavy research writeups. |\n| `@walterra/pi-charts` | Charts and quantitative visualizations. |\n| `pi-mermaid` | Diagrams in the TUI. |\n| `@aliou/pi-processes` | Long-running experiments and log tails. |\n| `pi-zotero` | Citation-library workflows. |\n| `pi-schedule-prompt` | Recurring and deferred research jobs. |\n| `@tmustier/pi-ralph-wiggum` | Long-running agent loops for iterative development. |\n\n## Optional packages\n\nInstall on demand with `feynman packages install \u003Cpreset>`.\n\n| Package | Purpose |\n|---------|---------|\n| `pi-generative-ui` | Interactive HTML-style widgets. |\n| `@kaiserlich-dev/pi-session-search` | Indexed session recall and summarize/resume UI. |\n| `@samfp/pi-memory` | Automatic preference and correction memory across sessions. |","src/content/docs/reference/package-stack.md","7c0accc036d20333",{"html":307,"metadata":308},"\u003Cp>Curated Pi packages bundled with Feynman. The runtime package list lives in \u003Ccode>.feynman/settings.json\u003C/code>.\u003C/p>\n\u003Ch2 id=\"core-packages\">Core packages\u003C/h2>\n\u003Cp>Installed by default.\u003C/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Package\u003C/th>\u003Cth>Purpose\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>pi-subagents\u003C/code>\u003C/td>\u003Ctd>Parallel literature gathering and decomposition.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>pi-btw\u003C/code>\u003C/td>\u003Ctd>Fast side-thread \u003Ccode>/btw\u003C/code> conversations without interrupting the main run.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>pi-docparser\u003C/code>\u003C/td>\u003Ctd>PDFs, Office docs, spreadsheets, and images.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>pi-web-access\u003C/code>\u003C/td>\u003Ctd>Web, GitHub, PDF, and media access.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>pi-markdown-preview\u003C/code>\u003C/td>\u003Ctd>Polished Markdown and LaTeX-heavy research writeups.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>@walterra/pi-charts\u003C/code>\u003C/td>\u003Ctd>Charts and quantitative visualizations.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>pi-mermaid\u003C/code>\u003C/td>\u003Ctd>Diagrams in the TUI.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>@aliou/pi-processes\u003C/code>\u003C/td>\u003Ctd>Long-running experiments and log tails.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>pi-zotero\u003C/code>\u003C/td>\u003Ctd>Citation-library workflows.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>pi-schedule-prompt\u003C/code>\u003C/td>\u003Ctd>Recurring and deferred research jobs.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>@tmustier/pi-ralph-wiggum\u003C/code>\u003C/td>\u003Ctd>Long-running agent loops for iterative development.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch2 id=\"optional-packages\">Optional packages\u003C/h2>\n\u003Cp>Install on demand with \u003Ccode>feynman packages install <preset>\u003C/code>.\u003C/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Package\u003C/th>\u003Cth>Purpose\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>pi-generative-ui\u003C/code>\u003C/td>\u003Ctd>Interactive HTML-style widgets.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>@kaiserlich-dev/pi-session-search\u003C/code>\u003C/td>\u003Ctd>Indexed session recall and summarize/resume UI.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>@samfp/pi-memory\u003C/code>\u003C/td>\u003Ctd>Automatic preference and correction memory across sessions.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>",{"headings":309,"localImagePaths":314,"remoteImagePaths":315,"frontmatter":316,"imagePaths":317},[310,313],{"depth":17,"slug":311,"text":312},"core-packages","Core packages",{"depth":17,"slug":161,"text":162},[],[],{"title":301,"description":302,"section":241,"order":41},[],"reference/package-stack.md","tools/session-search",{"id":319,"data":321,"body":324,"filePath":325,"digest":326,"rendered":327,"legacyId":344},{"title":322,"description":323,"section":74,"order":41},"Session Search","Search prior Feynman session transcripts","## Overview\n\nThe `session_search` tool recovers prior Feynman work from stored session transcripts. Useful for picking up previous research threads or finding past findings.\n\n## Usage\n\nInside the REPL:\n\n```\n/search\n```\n\nOr use the tool directly — Feynman will invoke `session_search` automatically when you reference prior work.\n\n## What it searches\n\n- Full session transcripts\n- Tool outputs and agent results\n- Generated artifacts and their content","src/content/docs/tools/session-search.md","7091dddc6969e581",{"html":328,"metadata":329},"\u003Ch2 id=\"overview\">Overview\u003C/h2>\n\u003Cp>The \u003Ccode>session_search\u003C/code> tool recovers prior Feynman work from stored session transcripts. Useful for picking up previous research threads or finding past findings.\u003C/p>\n\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cp>Inside the REPL:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/search\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>Or use the tool directly — Feynman will invoke \u003Ccode>session_search\u003C/code> automatically when you reference prior work.\u003C/p>\n\u003Ch2 id=\"what-it-searches\">What it searches\u003C/h2>\n\u003Cul>\n\u003Cli>Full session transcripts\u003C/li>\n\u003Cli>Tool outputs and agent results\u003C/li>\n\u003Cli>Generated artifacts and their content\u003C/li>\n\u003C/ul>",{"headings":330,"localImagePaths":340,"remoteImagePaths":341,"frontmatter":342,"imagePaths":343},[331,334,337],{"depth":17,"slug":332,"text":333},"overview","Overview",{"depth":17,"slug":335,"text":336},"usage","Usage",{"depth":17,"slug":338,"text":339},"what-it-searches","What it searches",[],[],{"title":322,"description":323,"section":74,"order":41},[],"tools/session-search.md","tools/alphaxiv",{"id":345,"data":347,"body":349,"filePath":350,"digest":351,"rendered":352,"legacyId":374},{"title":257,"description":348,"section":74,"order":94},"Paper search and analysis tools","## Overview\n\nAlphaXiv powers Feynman's academic paper workflows. All tools require an alphaXiv account — sign in with `feynman alpha login`.\n\n## Tools\n\n### alpha_search\n\nPaper discovery with three search modes:\n\n- **semantic** — Meaning-based search across paper content\n- **keyword** — Traditional keyword matching\n- **agentic** — AI-powered search that interprets your intent\n\n### alpha_get_paper\n\nFetch a paper's report (structured summary) or full raw text by arXiv ID.\n\n### alpha_ask_paper\n\nAsk a targeted question about a specific paper. Returns an answer grounded in the paper's content.\n\n### alpha_annotate_paper\n\nAdd persistent local notes to a paper. Annotations are stored locally and persist across sessions.\n\n### alpha_list_annotations\n\nRecall all annotations across papers and sessions.\n\n### alpha_read_code\n\nRead source code from a paper's linked GitHub repository. Useful for auditing or replication planning.","src/content/docs/tools/alphaxiv.md","a6eeb2c5a98d3096",{"html":353,"metadata":354},"\u003Ch2 id=\"overview\">Overview\u003C/h2>\n\u003Cp>AlphaXiv powers Feynman’s academic paper workflows. All tools require an alphaXiv account — sign in with \u003Ccode>feynman alpha login\u003C/code>.\u003C/p>\n\u003Ch2 id=\"tools\">Tools\u003C/h2>\n\u003Ch3 id=\"alpha_search\">alpha_search\u003C/h3>\n\u003Cp>Paper discovery with three search modes:\u003C/p>\n\u003Cul>\n\u003Cli>\u003Cstrong>semantic\u003C/strong> — Meaning-based search across paper content\u003C/li>\n\u003Cli>\u003Cstrong>keyword\u003C/strong> — Traditional keyword matching\u003C/li>\n\u003Cli>\u003Cstrong>agentic\u003C/strong> — AI-powered search that interprets your intent\u003C/li>\n\u003C/ul>\n\u003Ch3 id=\"alpha_get_paper\">alpha_get_paper\u003C/h3>\n\u003Cp>Fetch a paper’s report (structured summary) or full raw text by arXiv ID.\u003C/p>\n\u003Ch3 id=\"alpha_ask_paper\">alpha_ask_paper\u003C/h3>\n\u003Cp>Ask a targeted question about a specific paper. Returns an answer grounded in the paper’s content.\u003C/p>\n\u003Ch3 id=\"alpha_annotate_paper\">alpha_annotate_paper\u003C/h3>\n\u003Cp>Add persistent local notes to a paper. Annotations are stored locally and persist across sessions.\u003C/p>\n\u003Ch3 id=\"alpha_list_annotations\">alpha_list_annotations\u003C/h3>\n\u003Cp>Recall all annotations across papers and sessions.\u003C/p>\n\u003Ch3 id=\"alpha_read_code\">alpha_read_code\u003C/h3>\n\u003Cp>Read source code from a paper’s linked GitHub repository. Useful for auditing or replication planning.\u003C/p>",{"headings":355,"localImagePaths":370,"remoteImagePaths":371,"frontmatter":372,"imagePaths":373},[356,357,358,360,362,364,366,368],{"depth":17,"slug":332,"text":333},{"depth":17,"slug":73,"text":74},{"depth":41,"slug":359,"text":359},"alpha_search",{"depth":41,"slug":361,"text":361},"alpha_get_paper",{"depth":41,"slug":363,"text":363},"alpha_ask_paper",{"depth":41,"slug":365,"text":365},"alpha_annotate_paper",{"depth":41,"slug":367,"text":367},"alpha_list_annotations",{"depth":41,"slug":369,"text":369},"alpha_read_code",[],[],{"title":257,"description":348,"section":74,"order":94},[],"tools/alphaxiv.md","tools/preview",{"id":375,"data":377,"body":380,"filePath":381,"digest":382,"rendered":383,"legacyId":397},{"title":378,"description":379,"section":74,"order":205},"Preview","Preview generated artifacts in browser or PDF","## Overview\n\nThe `preview_file` tool opens generated artifacts in your browser or PDF viewer.\n\n## Usage\n\nInside the REPL:\n\n```\n/preview\n```\n\nOr Feynman will suggest previewing when you generate artifacts that benefit from rendered output (Markdown with LaTeX, HTML reports, etc.).\n\n## Requirements\n\nPreview requires `pandoc` for PDF/HTML rendering. Install it with:\n\n```bash\nfeynman --setup-preview\n```\n\n## Supported formats\n\n- Markdown (with LaTeX math rendering)\n- HTML\n- PDF","src/content/docs/tools/preview.md","b42137d5e0befd83",{"html":384,"metadata":385},"\u003Ch2 id=\"overview\">Overview\u003C/h2>\n\u003Cp>The \u003Ccode>preview_file\u003C/code> tool opens generated artifacts in your browser or PDF viewer.\u003C/p>\n\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cp>Inside the REPL:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/preview\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>Or Feynman will suggest previewing when you generate artifacts that benefit from rendered output (Markdown with LaTeX, HTML reports, etc.).\u003C/p>\n\u003Ch2 id=\"requirements\">Requirements\u003C/h2>\n\u003Cp>Preview requires \u003Ccode>pandoc\u003C/code> for PDF/HTML rendering. Install it with:\u003C/p>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#005CC5;--shiki-dark:#79B8FF\"> --setup-preview\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"supported-formats\">Supported formats\u003C/h2>\n\u003Cul>\n\u003Cli>Markdown (with LaTeX math rendering)\u003C/li>\n\u003Cli>HTML\u003C/li>\n\u003Cli>PDF\u003C/li>\n\u003C/ul>",{"headings":386,"localImagePaths":393,"remoteImagePaths":394,"frontmatter":395,"imagePaths":396},[387,388,389,390],{"depth":17,"slug":332,"text":333},{"depth":17,"slug":335,"text":336},{"depth":17,"slug":282,"text":283},{"depth":17,"slug":391,"text":392},"supported-formats","Supported formats",[],[],{"title":378,"description":379,"section":74,"order":205},[],"tools/preview.md","reference/slash-commands",{"id":398,"data":400,"body":403,"filePath":404,"digest":405,"rendered":406,"legacyId":422},{"title":401,"description":402,"section":241,"order":17},"Slash Commands","Repo-owned REPL slash commands","This page documents the slash commands that Feynman owns in this repository: prompt templates from `prompts/` and extension commands from `extensions/research-tools/`.\n\nAdditional slash commands can appear at runtime from Pi core and bundled packages such as subagents, preview, session search, and scheduling. Use `/help` inside the REPL for the live command list instead of relying on a static copy of package-provided commands.\n\n## Research Workflows\n\n| Command | Description |\n| --- | --- |\n| `/deepresearch \u003Ctopic>` | Run a thorough, source-heavy investigation on a topic and produce a durable research brief with inline citations. |\n| `/lit \u003Ctopic>` | Run a literature review on a topic using paper search and primary-source synthesis. |\n| `/review \u003Cartifact>` | Simulate an AI research peer review with likely objections, severity, and a concrete revision plan. |\n| `/audit \u003Citem>` | Compare a paper's claims against its public codebase and identify mismatches, omissions, and reproducibility risks. |\n| `/replicate \u003Cpaper>` | Plan or execute a replication workflow for a paper, claim, or benchmark. |\n| `/compare \u003Ctopic>` | Compare multiple sources on a topic and produce a source-grounded matrix of agreements, disagreements, and confidence. |\n| `/draft \u003Ctopic>` | Turn research findings into a polished paper-style draft with equations, sections, and explicit claims. |\n| `/autoresearch \u003Cidea>` | Autonomous experiment loop — try ideas, measure results, keep what works, discard what doesn't, repeat. |\n| `/watch \u003Ctopic>` | Set up a recurring or deferred research watch on a topic, company, paper area, or product surface. |\n\n## Project & Session\n\n| Command | Description |\n| --- | --- |\n| `/log` | Write a durable session log with completed work, findings, open questions, and next steps. |\n| `/jobs` | Inspect active background research work, including running processes and scheduled follow-ups. |\n| `/help` | Show grouped Feynman commands and prefill the editor with a selected command. |\n| `/init` | Bootstrap AGENTS.md and session-log folders for a research project. |\n\n## Setup\n\n| Command | Description |\n| --- | --- |\n| `/alpha-login` | Sign in to alphaXiv from inside Feynman. |\n| `/alpha-status` | Show alphaXiv authentication status. |\n| `/alpha-logout` | Clear alphaXiv auth from inside Feynman. |","src/content/docs/reference/slash-commands.md","f548c25cfafb9aea",{"html":407,"metadata":408},"\u003Cp>This page documents the slash commands that Feynman owns in this repository: prompt templates from \u003Ccode>prompts/\u003C/code> and extension commands from \u003Ccode>extensions/research-tools/\u003C/code>.\u003C/p>\n\u003Cp>Additional slash commands can appear at runtime from Pi core and bundled packages such as subagents, preview, session search, and scheduling. Use \u003Ccode>/help\u003C/code> inside the REPL for the live command list instead of relying on a static copy of package-provided commands.\u003C/p>\n\u003Ch2 id=\"research-workflows\">Research Workflows\u003C/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Command\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>/deepresearch <topic>\u003C/code>\u003C/td>\u003Ctd>Run a thorough, source-heavy investigation on a topic and produce a durable research brief with inline citations.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/lit <topic>\u003C/code>\u003C/td>\u003Ctd>Run a literature review on a topic using paper search and primary-source synthesis.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/review <artifact>\u003C/code>\u003C/td>\u003Ctd>Simulate an AI research peer review with likely objections, severity, and a concrete revision plan.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/audit <item>\u003C/code>\u003C/td>\u003Ctd>Compare a paper’s claims against its public codebase and identify mismatches, omissions, and reproducibility risks.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/replicate <paper>\u003C/code>\u003C/td>\u003Ctd>Plan or execute a replication workflow for a paper, claim, or benchmark.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/compare <topic>\u003C/code>\u003C/td>\u003Ctd>Compare multiple sources on a topic and produce a source-grounded matrix of agreements, disagreements, and confidence.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/draft <topic>\u003C/code>\u003C/td>\u003Ctd>Turn research findings into a polished paper-style draft with equations, sections, and explicit claims.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/autoresearch <idea>\u003C/code>\u003C/td>\u003Ctd>Autonomous experiment loop — try ideas, measure results, keep what works, discard what doesn’t, repeat.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/watch <topic>\u003C/code>\u003C/td>\u003Ctd>Set up a recurring or deferred research watch on a topic, company, paper area, or product surface.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch2 id=\"project--session\">Project & Session\u003C/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Command\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>/log\u003C/code>\u003C/td>\u003Ctd>Write a durable session log with completed work, findings, open questions, and next steps.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/jobs\u003C/code>\u003C/td>\u003Ctd>Inspect active background research work, including running processes and scheduled follow-ups.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/help\u003C/code>\u003C/td>\u003Ctd>Show grouped Feynman commands and prefill the editor with a selected command.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/init\u003C/code>\u003C/td>\u003Ctd>Bootstrap AGENTS.md and session-log folders for a research project.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch2 id=\"setup\">Setup\u003C/h2>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Command\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>/alpha-login\u003C/code>\u003C/td>\u003Ctd>Sign in to alphaXiv from inside Feynman.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/alpha-status\u003C/code>\u003C/td>\u003Ctd>Show alphaXiv authentication status.\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>/alpha-logout\u003C/code>\u003C/td>\u003Ctd>Clear alphaXiv auth from inside Feynman.\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>",{"headings":409,"localImagePaths":418,"remoteImagePaths":419,"frontmatter":420,"imagePaths":421},[410,413,416],{"depth":17,"slug":411,"text":412},"research-workflows","Research Workflows",{"depth":17,"slug":414,"text":415},"project--session","Project & Session",{"depth":17,"slug":417,"text":135},"setup",[],[],{"title":401,"description":402,"section":241,"order":17},[],"reference/slash-commands.md","tools/web-search",{"id":423,"data":425,"body":428,"filePath":429,"digest":430,"rendered":431,"legacyId":451},{"title":426,"description":427,"section":74,"order":17},"Web Search","Web search routing and configuration","## Routing modes\n\nFeynman supports three web search backends:\n\n| Mode | Description |\n|------|-------------|\n| `auto` | Prefer Perplexity when configured, fall back to Gemini |\n| `perplexity` | Force Perplexity Sonar |\n| `gemini` | Force Gemini (default) |\n\n## Default behavior\n\nThe default path is zero-config Gemini Browser via a signed-in Chromium profile. No API keys required.\n\n## Check current config\n\n```bash\nfeynman search status\n```\n\n## Advanced configuration\n\nEdit `~/.feynman/web-search.json` directly to set:\n\n- Gemini API keys\n- Perplexity API keys\n- Custom routing preferences","src/content/docs/tools/web-search.md","b2963fe8f7ae5dce",{"html":432,"metadata":433},"\u003Ch2 id=\"routing-modes\">Routing modes\u003C/h2>\n\u003Cp>Feynman supports three web search backends:\u003C/p>\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\u003Ctable>\u003Cthead>\u003Ctr>\u003Cth>Mode\u003C/th>\u003Cth>Description\u003C/th>\u003C/tr>\u003C/thead>\u003Ctbody>\u003Ctr>\u003Ctd>\u003Ccode>auto\u003C/code>\u003C/td>\u003Ctd>Prefer Perplexity when configured, fall back to Gemini\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>perplexity\u003C/code>\u003C/td>\u003Ctd>Force Perplexity Sonar\u003C/td>\u003C/tr>\u003Ctr>\u003Ctd>\u003Ccode>gemini\u003C/code>\u003C/td>\u003Ctd>Force Gemini (default)\u003C/td>\u003C/tr>\u003C/tbody>\u003C/table>\n\u003Ch2 id=\"default-behavior\">Default behavior\u003C/h2>\n\u003Cp>The default path is zero-config Gemini Browser via a signed-in Chromium profile. No API keys required.\u003C/p>\n\u003Ch2 id=\"check-current-config\">Check current config\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"bash\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan style=\"color:#6F42C1;--shiki-dark:#B392F0\">feynman\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> search\u003C/span>\u003Cspan style=\"color:#032F62;--shiki-dark:#9ECBFF\"> status\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"advanced-configuration\">Advanced configuration\u003C/h2>\n\u003Cp>Edit \u003Ccode>~/.feynman/web-search.json\u003C/code> directly to set:\u003C/p>\n\u003Cul>\n\u003Cli>Gemini API keys\u003C/li>\n\u003Cli>Perplexity API keys\u003C/li>\n\u003Cli>Custom routing preferences\u003C/li>\n\u003C/ul>",{"headings":434,"localImagePaths":447,"remoteImagePaths":448,"frontmatter":449,"imagePaths":450},[435,438,441,444],{"depth":17,"slug":436,"text":437},"routing-modes","Routing modes",{"depth":17,"slug":439,"text":440},"default-behavior","Default behavior",{"depth":17,"slug":442,"text":443},"check-current-config","Check current config",{"depth":17,"slug":445,"text":446},"advanced-configuration","Advanced configuration",[],[],{"title":426,"description":427,"section":74,"order":17},[],"tools/web-search.md","workflows/autoresearch",{"id":452,"data":454,"body":459,"filePath":460,"digest":461,"rendered":462,"legacyId":483},{"title":455,"description":456,"section":457,"order":458},"Autoresearch","Autonomous experiment optimization loop","Workflows",8,"## Usage\n\n```\n/autoresearch \u003Cidea>\n```\n\n## What it does\n\nRuns an autonomous experiment loop:\n\n1. **Edit** — Modify code or configuration\n2. **Commit** — Save the change\n3. **Benchmark** — Run evaluation\n4. **Evaluate** — Compare against baseline\n5. **Keep or revert** — Persist improvements, roll back regressions\n6. **Repeat** — Continue until the target is hit\n\n## Tracking\n\nMetrics are tracked in:\n\n- `autoresearch.md` — Human-readable progress log\n- `autoresearch.jsonl` — Machine-readable metrics over time\n\n## Controls\n\n```\n/autoresearch \u003Cidea> # start or resume\n/autoresearch off # stop, keep data\n/autoresearch clear # delete all state, start fresh\n```\n\n## Example\n\n```\n/autoresearch optimize the learning rate schedule for better convergence\n```","src/content/docs/workflows/autoresearch.md","94559e14e60edcad",{"html":463,"metadata":464},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/autoresearch <idea>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Runs an autonomous experiment loop:\u003C/p>\n\u003Col>\n\u003Cli>\u003Cstrong>Edit\u003C/strong> — Modify code or configuration\u003C/li>\n\u003Cli>\u003Cstrong>Commit\u003C/strong> — Save the change\u003C/li>\n\u003Cli>\u003Cstrong>Benchmark\u003C/strong> — Run evaluation\u003C/li>\n\u003Cli>\u003Cstrong>Evaluate\u003C/strong> — Compare against baseline\u003C/li>\n\u003Cli>\u003Cstrong>Keep or revert\u003C/strong> — Persist improvements, roll back regressions\u003C/li>\n\u003Cli>\u003Cstrong>Repeat\u003C/strong> — Continue until the target is hit\u003C/li>\n\u003C/ol>\n\u003Ch2 id=\"tracking\">Tracking\u003C/h2>\n\u003Cp>Metrics are tracked in:\u003C/p>\n\u003Cul>\n\u003Cli>\u003Ccode>autoresearch.md\u003C/code> — Human-readable progress log\u003C/li>\n\u003Cli>\u003Ccode>autoresearch.jsonl\u003C/code> — Machine-readable metrics over time\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"controls\">Controls\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/autoresearch <idea> # start or resume\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan>/autoresearch off # stop, keep data\u003C/span>\u003C/span>\n\u003Cspan class=\"line\">\u003Cspan>/autoresearch clear # delete all state, start fresh\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/autoresearch optimize the learning rate schedule for better convergence\u003C/span>\u003C/span>\u003C/code>\u003C/pre>",{"headings":465,"localImagePaths":479,"remoteImagePaths":480,"frontmatter":481,"imagePaths":482},[466,467,470,473,476],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},"what-it-does","What it does",{"depth":17,"slug":471,"text":472},"tracking","Tracking",{"depth":17,"slug":474,"text":475},"controls","Controls",{"depth":17,"slug":477,"text":478},"example","Example",[],[],{"title":455,"description":456,"section":457,"order":458},[],"workflows/autoresearch.md","workflows/audit",{"id":484,"data":486,"body":489,"filePath":490,"digest":491,"rendered":492,"legacyId":509},{"title":487,"description":488,"section":457,"order":205},"Code Audit","Compare paper claims against public codebases","## Usage\n\n```\n/audit \u003Citem>\n```\n\n## What it does\n\nCompares claims made in a paper against its public codebase. Surfaces mismatches, missing experiments, and reproducibility risks.\n\n## What it checks\n\n- Do the reported hyperparameters match the code?\n- Are all claimed experiments present in the repository?\n- Does the training loop match the described methodology?\n- Are there undocumented preprocessing steps?\n- Do evaluation metrics match the paper's claims?\n\n## Example\n\n```\n/audit 2401.12345\n```\n\n## Output\n\nAn audit report with:\n\n- Claim-by-claim verification\n- Identified mismatches\n- Missing components\n- Reproducibility risk assessment","src/content/docs/workflows/audit.md","58f5516850bcd065",{"html":493,"metadata":494},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/audit <item>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Compares claims made in a paper against its public codebase. Surfaces mismatches, missing experiments, and reproducibility risks.\u003C/p>\n\u003Ch2 id=\"what-it-checks\">What it checks\u003C/h2>\n\u003Cul>\n\u003Cli>Do the reported hyperparameters match the code?\u003C/li>\n\u003Cli>Are all claimed experiments present in the repository?\u003C/li>\n\u003Cli>Does the training loop match the described methodology?\u003C/li>\n\u003Cli>Are there undocumented preprocessing steps?\u003C/li>\n\u003Cli>Do evaluation metrics match the paper’s claims?\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/audit 2401.12345\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"output\">Output\u003C/h2>\n\u003Cp>An audit report with:\u003C/p>\n\u003Cul>\n\u003Cli>Claim-by-claim verification\u003C/li>\n\u003Cli>Identified mismatches\u003C/li>\n\u003Cli>Missing components\u003C/li>\n\u003Cli>Reproducibility risk assessment\u003C/li>\n\u003C/ul>",{"headings":495,"localImagePaths":505,"remoteImagePaths":506,"frontmatter":507,"imagePaths":508},[496,497,498,501,502],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},{"depth":17,"slug":499,"text":500},"what-it-checks","What it checks",{"depth":17,"slug":477,"text":478},{"depth":17,"slug":503,"text":504},"output","Output",[],[],{"title":487,"description":488,"section":457,"order":205},[],"workflows/audit.md","workflows/compare",{"id":510,"data":512,"body":516,"filePath":517,"digest":518,"rendered":519,"legacyId":531},{"title":513,"description":514,"section":457,"order":515},"Source Comparison","Compare multiple sources with agreement/disagreement matrix",6,"## Usage\n\n```\n/compare \u003Ctopic>\n```\n\n## What it does\n\nCompares multiple sources on a topic. Builds an agreement/disagreement matrix showing where sources align and where they conflict.\n\n## Example\n\n```\n/compare approaches to constitutional AI training\n```\n\n## Output\n\n- Source-by-source breakdown\n- Agreement/disagreement matrix\n- Synthesis of key differences\n- Assessment of which positions have stronger evidence","src/content/docs/workflows/compare.md","669d1dce304b191f",{"html":520,"metadata":521},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/compare <topic>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Compares multiple sources on a topic. Builds an agreement/disagreement matrix showing where sources align and where they conflict.\u003C/p>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/compare approaches to constitutional AI training\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"output\">Output\u003C/h2>\n\u003Cul>\n\u003Cli>Source-by-source breakdown\u003C/li>\n\u003Cli>Agreement/disagreement matrix\u003C/li>\n\u003Cli>Synthesis of key differences\u003C/li>\n\u003Cli>Assessment of which positions have stronger evidence\u003C/li>\n\u003C/ul>",{"headings":522,"localImagePaths":527,"remoteImagePaths":528,"frontmatter":529,"imagePaths":530},[523,524,525,526],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},{"depth":17,"slug":477,"text":478},{"depth":17,"slug":503,"text":504},[],[],{"title":513,"description":514,"section":457,"order":515},[],"workflows/compare.md","workflows/deep-research",{"id":532,"data":534,"body":537,"filePath":538,"digest":539,"rendered":540,"legacyId":552},{"title":535,"description":536,"section":457,"order":94},"Deep Research","Thorough source-heavy investigation with parallel agents","## Usage\n\n```\n/deepresearch \u003Ctopic>\n```\n\n## What it does\n\nDeep research runs a thorough, source-heavy investigation. It plans the research scope, delegates to parallel researcher agents, synthesizes findings, and adds inline citations.\n\nThe workflow follows these steps:\n\n1. **Plan** — Clarify the research question and identify search strategy\n2. **Delegate** — Spawn parallel researcher agents to gather evidence from different source types (papers, web, repos)\n3. **Synthesize** — Merge findings, resolve contradictions, identify gaps\n4. **Cite** — Add inline citations and verify all source URLs\n5. **Deliver** — Write a durable research brief to `outputs/`\n\n## Example\n\n```\n/deepresearch transformer scaling laws and their implications for compute-optimal training\n```\n\n## Output\n\nProduces a structured research brief with:\n\n- Executive summary\n- Key findings organized by theme\n- Evidence tables with source links\n- Open questions and suggested next steps\n- Numbered sources section with direct URLs","src/content/docs/workflows/deep-research.md","5a1ed5d3fd031659",{"html":541,"metadata":542},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/deepresearch <topic>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Deep research runs a thorough, source-heavy investigation. It plans the research scope, delegates to parallel researcher agents, synthesizes findings, and adds inline citations.\u003C/p>\n\u003Cp>The workflow follows these steps:\u003C/p>\n\u003Col>\n\u003Cli>\u003Cstrong>Plan\u003C/strong> — Clarify the research question and identify search strategy\u003C/li>\n\u003Cli>\u003Cstrong>Delegate\u003C/strong> — Spawn parallel researcher agents to gather evidence from different source types (papers, web, repos)\u003C/li>\n\u003Cli>\u003Cstrong>Synthesize\u003C/strong> — Merge findings, resolve contradictions, identify gaps\u003C/li>\n\u003Cli>\u003Cstrong>Cite\u003C/strong> — Add inline citations and verify all source URLs\u003C/li>\n\u003Cli>\u003Cstrong>Deliver\u003C/strong> — Write a durable research brief to \u003Ccode>outputs/\u003C/code>\u003C/li>\n\u003C/ol>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/deepresearch transformer scaling laws and their implications for compute-optimal training\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"output\">Output\u003C/h2>\n\u003Cp>Produces a structured research brief with:\u003C/p>\n\u003Cul>\n\u003Cli>Executive summary\u003C/li>\n\u003Cli>Key findings organized by theme\u003C/li>\n\u003Cli>Evidence tables with source links\u003C/li>\n\u003Cli>Open questions and suggested next steps\u003C/li>\n\u003Cli>Numbered sources section with direct URLs\u003C/li>\n\u003C/ul>",{"headings":543,"localImagePaths":548,"remoteImagePaths":549,"frontmatter":550,"imagePaths":551},[544,545,546,547],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},{"depth":17,"slug":477,"text":478},{"depth":17,"slug":503,"text":504},[],[],{"title":535,"description":536,"section":457,"order":94},[],"workflows/deep-research.md","workflows/draft",{"id":553,"data":555,"body":559,"filePath":560,"digest":561,"rendered":562,"legacyId":576},{"title":556,"description":557,"section":457,"order":558},"Draft Writing","Paper-style draft generation from research findings",7,"## Usage\n\n```\n/draft \u003Ctopic>\n```\n\n## What it does\n\nProduces a paper-style draft with structured sections. Writes to `papers/`.\n\n## Structure\n\nThe generated draft includes:\n\n- Title\n- Abstract\n- Introduction / Background\n- Method or Approach\n- Evidence and Analysis\n- Limitations\n- Conclusion\n- Sources\n\n## Example\n\n```\n/draft survey of differentiable physics simulators\n```\n\nThe writer agent works only from supplied evidence — it never fabricates content. If evidence is insufficient, it explicitly notes the gaps.","src/content/docs/workflows/draft.md","5549e489883745ea",{"html":563,"metadata":564},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/draft <topic>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Produces a paper-style draft with structured sections. Writes to \u003Ccode>papers/\u003C/code>.\u003C/p>\n\u003Ch2 id=\"structure\">Structure\u003C/h2>\n\u003Cp>The generated draft includes:\u003C/p>\n\u003Cul>\n\u003Cli>Title\u003C/li>\n\u003Cli>Abstract\u003C/li>\n\u003Cli>Introduction / Background\u003C/li>\n\u003Cli>Method or Approach\u003C/li>\n\u003Cli>Evidence and Analysis\u003C/li>\n\u003Cli>Limitations\u003C/li>\n\u003Cli>Conclusion\u003C/li>\n\u003Cli>Sources\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/draft survey of differentiable physics simulators\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Cp>The writer agent works only from supplied evidence — it never fabricates content. If evidence is insufficient, it explicitly notes the gaps.\u003C/p>",{"headings":565,"localImagePaths":572,"remoteImagePaths":573,"frontmatter":574,"imagePaths":575},[566,567,568,571],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},{"depth":17,"slug":569,"text":570},"structure","Structure",{"depth":17,"slug":477,"text":478},[],[],{"title":556,"description":557,"section":457,"order":558},[],"workflows/draft.md","workflows/replication",{"id":577,"data":579,"body":583,"filePath":584,"digest":585,"rendered":586,"legacyId":598},{"title":580,"description":581,"section":457,"order":582},"Replication","Plan replications of papers and claims",5,"## Usage\n\n```\n/replicate \u003Cpaper or claim>\n```\n\n## What it does\n\nExtracts key implementation details from a paper, identifies what's needed to replicate the results, and asks where to run before executing anything.\n\nBefore running code, Feynman asks you to choose an execution environment:\n\n- **Local** — run in the current working directory\n- **Virtual environment** — create an isolated venv/conda env first\n- **Docker** — run experiment code inside an isolated Docker container\n- **Cloud** — delegate to a remote Agent Computer machine\n- **Plan only** — produce the replication plan without executing\n\n## Example\n\n```\n/replicate \"chain-of-thought prompting improves math reasoning\"\n```\n\n## Output\n\nA replication plan covering:\n\n- Key claims to verify\n- Required resources (compute, data, models)\n- Implementation details extracted from the paper\n- Potential pitfalls and underspecified details\n- Step-by-step replication procedure\n- Success criteria\n\nIf an execution environment is selected, also produces runnable scripts and captured results.","src/content/docs/workflows/replication.md","838b8fa26ebbe08d",{"html":587,"metadata":588},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/replicate <paper or claim>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Extracts key implementation details from a paper, identifies what’s needed to replicate the results, and asks where to run before executing anything.\u003C/p>\n\u003Cp>Before running code, Feynman asks you to choose an execution environment:\u003C/p>\n\u003Cul>\n\u003Cli>\u003Cstrong>Local\u003C/strong> — run in the current working directory\u003C/li>\n\u003Cli>\u003Cstrong>Virtual environment\u003C/strong> — create an isolated venv/conda env first\u003C/li>\n\u003Cli>\u003Cstrong>Docker\u003C/strong> — run experiment code inside an isolated Docker container\u003C/li>\n\u003Cli>\u003Cstrong>Cloud\u003C/strong> — delegate to a remote Agent Computer machine\u003C/li>\n\u003Cli>\u003Cstrong>Plan only\u003C/strong> — produce the replication plan without executing\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/replicate \"chain-of-thought prompting improves math reasoning\"\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"output\">Output\u003C/h2>\n\u003Cp>A replication plan covering:\u003C/p>\n\u003Cul>\n\u003Cli>Key claims to verify\u003C/li>\n\u003Cli>Required resources (compute, data, models)\u003C/li>\n\u003Cli>Implementation details extracted from the paper\u003C/li>\n\u003Cli>Potential pitfalls and underspecified details\u003C/li>\n\u003Cli>Step-by-step replication procedure\u003C/li>\n\u003Cli>Success criteria\u003C/li>\n\u003C/ul>\n\u003Cp>If an execution environment is selected, also produces runnable scripts and captured results.\u003C/p>",{"headings":589,"localImagePaths":594,"remoteImagePaths":595,"frontmatter":596,"imagePaths":597},[590,591,592,593],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},{"depth":17,"slug":477,"text":478},{"depth":17,"slug":503,"text":504},[],[],{"title":580,"description":581,"section":457,"order":582},[],"workflows/replication.md","workflows/review",{"id":599,"data":601,"body":604,"filePath":605,"digest":606,"rendered":607,"legacyId":622},{"title":602,"description":603,"section":457,"order":41},"Peer Review","Simulated peer review with severity-graded feedback","## Usage\n\n```\n/review \u003Cartifact>\n```\n\n## What it does\n\nSimulates a tough-but-fair peer review for AI research artifacts. Evaluates novelty, empirical rigor, baselines, ablations, and reproducibility.\n\nThe reviewer agent identifies:\n\n- Weak baselines\n- Missing ablations\n- Evaluation mismatches\n- Benchmark leakage\n- Under-specified implementation details\n\n## Severity levels\n\nFeedback is graded by severity:\n\n- **FATAL** — Fundamental issues that invalidate the claims\n- **MAJOR** — Significant problems that need addressing\n- **MINOR** — Small improvements or clarifications\n\n## Example\n\n```\n/review outputs/scaling-laws-brief.md\n```\n\n## Output\n\nStructured review with:\n\n- Summary of the work\n- Strengths\n- Weaknesses (severity-graded)\n- Questions for the authors\n- Verdict (accept / revise / reject)\n- Revision plan","src/content/docs/workflows/review.md","5a1cfb4bdd03056c",{"html":608,"metadata":609},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/review <artifact>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Simulates a tough-but-fair peer review for AI research artifacts. Evaluates novelty, empirical rigor, baselines, ablations, and reproducibility.\u003C/p>\n\u003Cp>The reviewer agent identifies:\u003C/p>\n\u003Cul>\n\u003Cli>Weak baselines\u003C/li>\n\u003Cli>Missing ablations\u003C/li>\n\u003Cli>Evaluation mismatches\u003C/li>\n\u003Cli>Benchmark leakage\u003C/li>\n\u003Cli>Under-specified implementation details\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"severity-levels\">Severity levels\u003C/h2>\n\u003Cp>Feedback is graded by severity:\u003C/p>\n\u003Cul>\n\u003Cli>\u003Cstrong>FATAL\u003C/strong> — Fundamental issues that invalidate the claims\u003C/li>\n\u003Cli>\u003Cstrong>MAJOR\u003C/strong> — Significant problems that need addressing\u003C/li>\n\u003Cli>\u003Cstrong>MINOR\u003C/strong> — Small improvements or clarifications\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/review outputs/scaling-laws-brief.md\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"output\">Output\u003C/h2>\n\u003Cp>Structured review with:\u003C/p>\n\u003Cul>\n\u003Cli>Summary of the work\u003C/li>\n\u003Cli>Strengths\u003C/li>\n\u003Cli>Weaknesses (severity-graded)\u003C/li>\n\u003Cli>Questions for the authors\u003C/li>\n\u003Cli>Verdict (accept / revise / reject)\u003C/li>\n\u003Cli>Revision plan\u003C/li>\n\u003C/ul>",{"headings":610,"localImagePaths":618,"remoteImagePaths":619,"frontmatter":620,"imagePaths":621},[611,612,613,616,617],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},{"depth":17,"slug":614,"text":615},"severity-levels","Severity levels",{"depth":17,"slug":477,"text":478},{"depth":17,"slug":503,"text":504},[],[],{"title":602,"description":603,"section":457,"order":41},[],"workflows/review.md","workflows/literature-review",{"id":623,"data":625,"body":628,"filePath":629,"digest":630,"rendered":631,"legacyId":643},{"title":626,"description":627,"section":457,"order":17},"Literature Review","Map consensus, disagreements, and open questions","## Usage\n\n```\n/lit \u003Ctopic>\n```\n\n## What it does\n\nRuns a structured literature review that searches across academic papers and web sources. Explicitly separates consensus findings from disagreements and open questions.\n\n## Example\n\n```\n/lit multimodal reasoning benchmarks for large language models\n```\n\n## Output\n\nA structured review covering:\n\n- **Consensus** — What the field agrees on\n- **Disagreements** — Where sources conflict\n- **Open questions** — What remains unresolved\n- **Sources** — Direct links to all referenced papers and articles","src/content/docs/workflows/literature-review.md","7def25e86b0bdc22",{"html":632,"metadata":633},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/lit <topic>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Runs a structured literature review that searches across academic papers and web sources. Explicitly separates consensus findings from disagreements and open questions.\u003C/p>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/lit multimodal reasoning benchmarks for large language models\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"output\">Output\u003C/h2>\n\u003Cp>A structured review covering:\u003C/p>\n\u003Cul>\n\u003Cli>\u003Cstrong>Consensus\u003C/strong> — What the field agrees on\u003C/li>\n\u003Cli>\u003Cstrong>Disagreements\u003C/strong> — Where sources conflict\u003C/li>\n\u003Cli>\u003Cstrong>Open questions\u003C/strong> — What remains unresolved\u003C/li>\n\u003Cli>\u003Cstrong>Sources\u003C/strong> — Direct links to all referenced papers and articles\u003C/li>\n\u003C/ul>",{"headings":634,"localImagePaths":639,"remoteImagePaths":640,"frontmatter":641,"imagePaths":642},[635,636,637,638],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},{"depth":17,"slug":477,"text":478},{"depth":17,"slug":503,"text":504},[],[],{"title":626,"description":627,"section":457,"order":17},[],"workflows/literature-review.md","workflows/watch",{"id":644,"data":646,"body":650,"filePath":651,"digest":652,"rendered":653,"legacyId":667},{"title":647,"description":648,"section":457,"order":649},"Watch","Recurring research monitoring",9,"## Usage\n\n```\n/watch \u003Ctopic>\n```\n\n## What it does\n\nSchedules a recurring research watch. Sets a baseline of current knowledge and defines what constitutes a meaningful change worth reporting.\n\n## Example\n\n```\n/watch new papers on test-time compute scaling\n```\n\n## How it works\n\n1. Feynman establishes a baseline by surveying current sources\n2. Defines change signals (new papers, updated results, new repos)\n3. Schedules periodic checks via `pi-schedule-prompt`\n4. Reports only when meaningful changes are detected","src/content/docs/workflows/watch.md","b24ebad68d8b9736",{"html":654,"metadata":655},"\u003Ch2 id=\"usage\">Usage\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/watch <topic>\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"what-it-does\">What it does\u003C/h2>\n\u003Cp>Schedules a recurring research watch. Sets a baseline of current knowledge and defines what constitutes a meaningful change worth reporting.\u003C/p>\n\u003Ch2 id=\"example\">Example\u003C/h2>\n\u003Cpre class=\"astro-code astro-code-themes github-light github-dark\" style=\"background-color:#fff;--shiki-dark-bg:#24292e;color:#24292e;--shiki-dark:#e1e4e8; overflow-x: auto;\" tabindex=\"0\" data-language=\"plaintext\">\u003Ccode>\u003Cspan class=\"line\">\u003Cspan>/watch new papers on test-time compute scaling\u003C/span>\u003C/span>\u003C/code>\u003C/pre>\n\u003Ch2 id=\"how-it-works\">How it works\u003C/h2>\n\u003Col>\n\u003Cli>Feynman establishes a baseline by surveying current sources\u003C/li>\n\u003Cli>Defines change signals (new papers, updated results, new repos)\u003C/li>\n\u003Cli>Schedules periodic checks via \u003Ccode>pi-schedule-prompt\u003C/code>\u003C/li>\n\u003Cli>Reports only when meaningful changes are detected\u003C/li>\n\u003C/ol>",{"headings":656,"localImagePaths":663,"remoteImagePaths":664,"frontmatter":665,"imagePaths":666},[657,658,659,660],{"depth":17,"slug":335,"text":336},{"depth":17,"slug":468,"text":469},{"depth":17,"slug":477,"text":478},{"depth":17,"slug":661,"text":662},"how-it-works","How it works",[],[],{"title":647,"description":648,"section":457,"order":649},[],"workflows/watch.md","agents/verifier",{"id":668,"data":670,"body":673,"filePath":674,"digest":675,"rendered":676,"legacyId":695},{"title":671,"description":672,"section":16,"order":205},"Verifier","Post-process a draft to add inline citations and verify every source URL.","## Source\n\nGenerated from `.feynman/agents/verifier.md`. Edit that prompt file, not this docs page.\n\n## Role\n\nPost-process a draft to add inline citations and verify every source URL.\n\n## Tools\n\n`read`, `bash`, `grep`, `find`, `ls`, `write`, `edit`\n\n## Default Output\n\n`cited.md`\n\nYou receive a draft document and the research files it was built from. Your job is to:\n\n1. **Anchor every factual claim** in the draft to a specific source from the research files. Insert inline citations `[1]`, `[2]`, etc. directly after each claim.\n2. **Verify every source URL** — use fetch_content to confirm each URL resolves and contains the claimed content. Flag dead links.\n3. **Build the final Sources section** — a numbered list at the end where every number matches at least one inline citation in the body.\n4. **Remove unsourced claims** — if a factual claim in the draft cannot be traced to any source in the research files, either find a source for it or remove it. Do not leave unsourced factual claims.\n\n## Citation rules\n\n- Every factual claim gets at least one citation: \"Transformers achieve 94.2% on MMLU [3].\"\n- Multiple sources for one claim: \"Recent work questions benchmark validity [7, 12].\"\n- No orphan citations — every `[N]` in the body must appear in Sources.\n- No orphan sources — every entry in Sources must be cited at least once.\n- Hedged or opinion statements do not need citations.\n- When multiple research files use different numbering, merge into a single unified sequence starting from [1]. Deduplicate sources that appear in multiple files.\n\n## Source verification\n\nFor each source URL:\n- **Live:** keep as-is.\n- **Dead/404:** search for an alternative URL (archived version, mirror, updated link). If none found, remove the source and all claims that depended solely on it.\n- **Redirects to unrelated content:** treat as dead.\n\n## Output contract\n- Save to the output file (default: `cited.md`).\n- The output is the complete final document — same structure as the input draft, but with inline citations added throughout and a verified Sources section.\n- Do not change the substance or structure of the draft. Only add citations and fix dead sources.","src/content/docs/agents/verifier.md","efc12a91a847824e",{"html":677,"metadata":678},"\u003Ch2 id=\"source\">Source\u003C/h2>\n\u003Cp>Generated from \u003Ccode>.feynman/agents/verifier.md\u003C/code>. Edit that prompt file, not this docs page.\u003C/p>\n\u003Ch2 id=\"role\">Role\u003C/h2>\n\u003Cp>Post-process a draft to add inline citations and verify every source URL.\u003C/p>\n\u003Ch2 id=\"tools\">Tools\u003C/h2>\n\u003Cp>\u003Ccode>read\u003C/code>, \u003Ccode>bash\u003C/code>, \u003Ccode>grep\u003C/code>, \u003Ccode>find\u003C/code>, \u003Ccode>ls\u003C/code>, \u003Ccode>write\u003C/code>, \u003Ccode>edit\u003C/code>\u003C/p>\n\u003Ch2 id=\"default-output\">Default Output\u003C/h2>\n\u003Cp>\u003Ccode>cited.md\u003C/code>\u003C/p>\n\u003Cp>You receive a draft document and the research files it was built from. Your job is to:\u003C/p>\n\u003Col>\n\u003Cli>\u003Cstrong>Anchor every factual claim\u003C/strong> in the draft to a specific source from the research files. Insert inline citations \u003Ccode>[1]\u003C/code>, \u003Ccode>[2]\u003C/code>, etc. directly after each claim.\u003C/li>\n\u003Cli>\u003Cstrong>Verify every source URL\u003C/strong> — use fetch_content to confirm each URL resolves and contains the claimed content. Flag dead links.\u003C/li>\n\u003Cli>\u003Cstrong>Build the final Sources section\u003C/strong> — a numbered list at the end where every number matches at least one inline citation in the body.\u003C/li>\n\u003Cli>\u003Cstrong>Remove unsourced claims\u003C/strong> — if a factual claim in the draft cannot be traced to any source in the research files, either find a source for it or remove it. Do not leave unsourced factual claims.\u003C/li>\n\u003C/ol>\n\u003Ch2 id=\"citation-rules\">Citation rules\u003C/h2>\n\u003Cul>\n\u003Cli>Every factual claim gets at least one citation: “Transformers achieve 94.2% on MMLU [3].”\u003C/li>\n\u003Cli>Multiple sources for one claim: “Recent work questions benchmark validity [7, 12].”\u003C/li>\n\u003Cli>No orphan citations — every \u003Ccode>[N]\u003C/code> in the body must appear in Sources.\u003C/li>\n\u003Cli>No orphan sources — every entry in Sources must be cited at least once.\u003C/li>\n\u003Cli>Hedged or opinion statements do not need citations.\u003C/li>\n\u003Cli>When multiple research files use different numbering, merge into a single unified sequence starting from [1]. Deduplicate sources that appear in multiple files.\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"source-verification\">Source verification\u003C/h2>\n\u003Cp>For each source URL:\u003C/p>\n\u003Cul>\n\u003Cli>\u003Cstrong>Live:\u003C/strong> keep as-is.\u003C/li>\n\u003Cli>\u003Cstrong>Dead/404:\u003C/strong> search for an alternative URL (archived version, mirror, updated link). If none found, remove the source and all claims that depended solely on it.\u003C/li>\n\u003Cli>\u003Cstrong>Redirects to unrelated content:\u003C/strong> treat as dead.\u003C/li>\n\u003C/ul>\n\u003Ch2 id=\"output-contract\">Output contract\u003C/h2>\n\u003Cul>\n\u003Cli>Save to the output file (default: \u003Ccode>cited.md\u003C/code>).\u003C/li>\n\u003Cli>The output is the complete final document — same structure as the input draft, but with inline citations added throughout and a verified Sources section.\u003C/li>\n\u003Cli>Do not change the substance or structure of the draft. Only add citations and fix dead sources.\u003C/li>\n\u003C/ul>",{"headings":679,"localImagePaths":691,"remoteImagePaths":692,"frontmatter":693,"imagePaths":694},[680,681,682,683,684,687,690],{"depth":17,"slug":26,"text":27},{"depth":17,"slug":29,"text":30},{"depth":17,"slug":73,"text":74},{"depth":17,"slug":32,"text":33},{"depth":17,"slug":685,"text":686},"citation-rules","Citation rules",{"depth":17,"slug":688,"text":689},"source-verification","Source verification",{"depth":17,"slug":51,"text":52},[],[],{"title":671,"description":672,"section":16,"order":205},[],"agents/verifier.md"] \ No newline at end of file diff --git a/website/src/content/docs/workflows/replication.md b/website/src/content/docs/workflows/replication.md index eaa97d1..c2ca4cf 100644 --- a/website/src/content/docs/workflows/replication.md +++ b/website/src/content/docs/workflows/replication.md @@ -20,7 +20,6 @@ Before running code, Feynman asks you to choose an execution environment: - **Local** — run in the current working directory - **Virtual environment** — create an isolated venv/conda env first - **Docker** — run experiment code inside an isolated Docker container -- **Cloud** — delegate to a remote Agent Computer machine - **Plan only** — produce the replication plan without executing ## Example diff --git a/website/src/pages/index.astro b/website/src/pages/index.astro index c79feeb..406461d 100644 --- a/website/src/pages/index.astro +++ b/website/src/pages/index.astro @@ -122,10 +122,6 @@ import AsciiLogo from '../components/AsciiLogo.astro';
Docker

Isolated container execution for safe local experiments

-
-
Agent Computer
-

Secure cloud execution for GPU workloads and long-running research

-
Web search

Gemini or Perplexity, zero-config default

@@ -144,7 +140,7 @@ import AsciiLogo from '../components/AsciiLogo.astro';
-

Built on Pi, alphaXiv, and Agent Computer. MIT licensed. Open source.

+

Built on Pi and alphaXiv. MIT licensed. Open source.

Get started GitHub