Refine Feynman research workflows

This commit is contained in:
Advait Paliwal
2026-03-22 12:19:33 -07:00
parent be97ac7a38
commit dd701e9967
24 changed files with 905 additions and 43 deletions

28
.pi/agents/auto.chain.md Normal file
View File

@@ -0,0 +1,28 @@
---
name: auto
description: Plan, investigate, verify, and draft an end-to-end autoresearch run.
---
## planner
output: plan.md
Clarify the objective, intended contribution, artifact, smallest useful experiment, and key open questions for {task}.
## researcher
reads: plan.md
output: research.md
Gather the strongest evidence, prior work, and concrete experiment options for {task} using plan.md as the scope guard.
## verifier
reads: plan.md+research.md
output: verification.md
Check whether the evidence and proposed claims for {task} are strong enough. Identify unsupported leaps, missing validation, and highest-value next checks.
## writer
reads: plan.md+research.md+verification.md
output: autoresearch.md
progress: true
Produce the final autoresearch artifact for {task}. If experiments were not run, be explicit about that. Preserve limitations and end with Sources.

22
.pi/agents/deep.chain.md Normal file
View File

@@ -0,0 +1,22 @@
---
name: deep
description: Gather, verify, and synthesize a deep research brief.
---
## researcher
output: research.md
Investigate {task}. Gather the strongest relevant primary sources, inspect them directly, and produce an evidence-first research brief.
## verifier
reads: research.md
output: verification.md
Verify the claims, source quality, and unresolved gaps in research.md for {task}. Produce a verification table and prioritized corrections.
## writer
reads: research.md+verification.md
output: deepresearch.md
progress: true
Write the final deep research brief for {task} using research.md and verification.md. Keep only supported claims, preserve caveats, and end with Sources.

28
.pi/agents/researcher.md Normal file
View File

@@ -0,0 +1,28 @@
---
name: researcher
description: Gather primary evidence across papers, web sources, repos, docs, and local artifacts.
thinking: high
output: research.md
defaultProgress: true
---
You are Feynman's evidence-gathering subagent.
Operating rules:
- Prefer primary sources: official docs, papers, datasets, repos, benchmarks, and direct experimental outputs.
- When the topic is current or market-facing, use web tools first; when it has literature depth, use paper tools as well.
- Do not rely on a single source type when the topic spans current reality and academic background.
- Inspect the strongest sources directly before summarizing them.
- Build a compact evidence table with:
- source
- key claim
- evidence type
- caveats
- confidence
- Preserve uncertainty explicitly and note disagreements across sources.
- Produce durable markdown that another agent can verify and another agent can turn into a polished artifact.
- End with a `Sources` section containing direct URLs.
Default output expectations:
- Save the main artifact to `research.md`.
- Keep it structured, terse, and evidence-first.

28
.pi/agents/verifier.md Normal file
View File

@@ -0,0 +1,28 @@
---
name: verifier
description: Verify claims, source quality, and evidentiary support in a research artifact.
thinking: high
output: verification.md
defaultProgress: true
---
You are Feynman's verification subagent.
Your job is to audit evidence, not to write a polished final narrative.
Operating rules:
- Check every strong claim against inspected sources or explicit experimental evidence.
- Label claims as:
- supported
- plausible inference
- disputed
- unsupported
- Look for stale sources, benchmark leakage, repo-paper mismatches, missing defaults, ambiguous methodology, and citation quality problems.
- Prefer precise corrections over broad rewrites.
- Produce a verification table plus a short prioritized list of fixes.
- Preserve open questions and unresolved disagreements instead of smoothing them away.
- End with a `Sources` section containing direct URLs for any additional material you inspected during verification.
Default output expectations:
- Save the main artifact to `verification.md`.
- Optimize for factual pressure-testing, not prose.

22
.pi/agents/writer.md Normal file
View File

@@ -0,0 +1,22 @@
---
name: writer
description: Turn verified research notes into clear memos, audits, and paper-style drafts.
thinking: medium
output: draft.md
defaultProgress: true
---
You are Feynman's writing subagent.
Operating rules:
- Write only from supplied evidence and clearly marked inference.
- Do not introduce unsupported claims.
- Preserve caveats, disagreements, and open questions instead of hiding them.
- Use clean Markdown structure and add equations only when they materially help.
- Keep the narrative readable, but never outrun the evidence.
- Produce artifacts that are ready to review in a browser or PDF preview.
- End with a `Sources` appendix containing direct URLs.
Default output expectations:
- Save the main artifact to `draft.md` unless the caller specifies a different output path.
- Optimize for clarity, structure, and evidence traceability.

View File

@@ -7,13 +7,15 @@
"paper2": "#1c1917",
"paper3": "#221f1c",
"panel": "#27231f",
"stone": "#9a9084",
"ash": "#6f675e",
"moss": "#24332c",
"moss2": "#202c26",
"stone": "#aaa79d",
"ash": "#909d91",
"darkAsh": "#4f4a44",
"oxide": "#b76e4c",
"gold": "#d0a85c",
"sage": "#88a88a",
"teal": "#7aa6a1",
"sage": "#86d8a4",
"teal": "#69d6c4",
"rose": "#c97b84",
"violet": "#a98dc6",
"selection": "#302b27",
@@ -21,31 +23,31 @@
"errorBg": "#2b1f21"
},
"colors": {
"accent": "gold",
"accent": "sage",
"border": "stone",
"borderAccent": "gold",
"borderAccent": "sage",
"borderMuted": "darkAsh",
"success": "sage",
"error": "rose",
"warning": "oxide",
"warning": "sage",
"muted": "stone",
"dim": "ash",
"text": "ink",
"thinkingText": "stone",
"thinkingText": "sage",
"selectedBg": "selection",
"userMessageBg": "panel",
"userMessageText": "ink",
"customMessageBg": "panel",
"customMessageText": "ink",
"customMessageLabel": "violet",
"userMessageBg": "moss",
"userMessageText": "",
"customMessageBg": "moss2",
"customMessageText": "",
"customMessageLabel": "sage",
"toolPendingBg": "paper2",
"toolSuccessBg": "successBg",
"toolErrorBg": "errorBg",
"toolTitle": "gold",
"toolTitle": "sage",
"toolOutput": "stone",
"mdHeading": "gold",
"mdHeading": "sage",
"mdLink": "teal",
"mdLinkUrl": "stone",
"mdCode": "teal",
@@ -54,30 +56,30 @@
"mdQuote": "stone",
"mdQuoteBorder": "ash",
"mdHr": "ash",
"mdListBullet": "gold",
"mdListBullet": "sage",
"toolDiffAdded": "sage",
"toolDiffRemoved": "rose",
"toolDiffContext": "stone",
"syntaxComment": "stone",
"syntaxKeyword": "gold",
"syntaxKeyword": "sage",
"syntaxFunction": "teal",
"syntaxVariable": "ink",
"syntaxString": "sage",
"syntaxNumber": "oxide",
"syntaxNumber": "teal",
"syntaxType": "violet",
"syntaxOperator": "ink",
"syntaxPunctuation": "stone",
"thinkingOff": "darkAsh",
"thinkingMinimal": "stone",
"thinkingMinimal": "ash",
"thinkingLow": "teal",
"thinkingMedium": "gold",
"thinkingMedium": "sage",
"thinkingHigh": "violet",
"thinkingXhigh": "rose",
"bashMode": "oxide"
"bashMode": "sage"
},
"export": {
"pageBg": "#141210",

View File

@@ -57,17 +57,32 @@ Most users should not need slash commands. The intended default is:
Inside the REPL:
- `/help` shows local commands
- `/init` bootstraps `AGENTS.md` and `notes/session-logs/`
- `/alpha-login` signs in to alphaXiv
- `/alpha-status` checks alphaXiv auth
- `/new` starts a new persisted session
- `/exit` quits
- `/lit-review <topic>` expands the literature-review prompt template
- `/lit <topic>` expands the literature-review prompt template
- `/replicate <paper or claim>` expands the replication prompt template
- `/reading-list <topic>` expands the reading-list prompt template
- `/research-memo <topic>` expands the general research memo prompt template
- `/reading <topic>` expands the reading-list prompt template
- `/memo <topic>` expands the general research memo prompt template
- `/deepresearch <topic>` expands the thorough source-heavy research prompt template
- `/autoresearch <idea>` expands the end-to-end idea-to-paper prompt template
- `/compare-sources <topic>` expands the source comparison prompt template
- `/paper-code-audit <item>` expands the paper/code audit prompt template
- `/paper-draft <topic>` expands the paper-style writing prompt template
- `/compare <topic>` expands the source comparison prompt template
- `/audit <item>` expands the paper/code audit prompt template
- `/draft <topic>` expands the paper-style writing prompt template
- `/log` writes a durable session log to `notes/`
- `/watch <topic>` schedules or prepares a recurring research watch
- `/jobs` inspects active background work
Package-powered workflows inside the REPL:
- `/agents` opens the subagent and chain manager
- `/run`, `/chain`, and `/parallel` delegate work to subagents
- `/ps` opens the background process panel
- `/schedule-prompt` manages recurring and deferred jobs
- `/search` opens indexed session search
- `/preview` previews generated artifacts in the terminal, browser, or PDF
Outside the REPL:
@@ -90,6 +105,14 @@ The starter extension adds:
- `session_search` for recovering prior Feynman work from stored transcripts
- `preview_file` for browser/PDF review of generated artifacts
Feynman also ships bundled research subagents in `.pi/agents/`:
- `researcher` for evidence gathering
- `verifier` for claim and source checking
- `writer` for polished memo and draft writing
- `deep` chain for gather → verify → synthesize
- `auto` chain for plan → gather → verify → draft
Feynman uses `@companion-ai/alpha-hub` directly in-process rather than shelling out to the CLI.
## Curated Pi Stack
@@ -115,6 +138,7 @@ The default expectation is source-grounded outputs with explicit `Sources` secti
```text
feynman/
├── .pi/agents/ # Bundled research subagents and chains
├── extensions/ # Custom research tools
├── papers/ # Polished paper-style drafts and writeups
├── prompts/ # Slash-style prompt templates

View File

@@ -1,4 +1,5 @@
import { execFile, spawn } from "node:child_process";
import { createRequire } from "node:module";
import { mkdir, mkdtemp, readFile, readdir, stat, writeFile } from "node:fs/promises";
import { homedir, tmpdir } from "node:os";
import { basename, dirname, extname, join, resolve as resolvePath } from "node:path";
@@ -10,7 +11,11 @@ import {
clearPaperAnnotation,
disconnect,
getPaper,
getUserName as getAlphaUserName,
isLoggedIn as isAlphaLoggedIn,
listPaperAnnotations,
login as loginAlpha,
logout as logoutAlpha,
readPaperCode,
searchPapers,
} from "@companion-ai/alpha-hub/lib";
@@ -18,6 +23,15 @@ import type { ExtensionAPI, ExtensionContext } from "@mariozechner/pi-coding-age
import { Type } from "@sinclair/typebox";
const execFileAsync = promisify(execFile);
const require = createRequire(import.meta.url);
const FEYNMAN_VERSION = (() => {
try {
const pkg = require("../package.json") as { version?: string };
return pkg.version ?? "dev";
} catch {
return "dev";
}
})();
function formatToolText(result: unknown): string {
return typeof result === "string" ? result : JSON.stringify(result, null, 2);
@@ -414,6 +428,170 @@ async function renderPdfPreview(filePath: string): Promise<string> {
return pdfPath;
}
function formatHeaderPath(path: string): string {
const home = homedir();
return path.startsWith(home) ? `~${path.slice(home.length)}` : path;
}
function truncateForWidth(text: string, width: number): string {
if (width <= 0) {
return "";
}
if (text.length <= width) {
return text;
}
if (width <= 3) {
return ".".repeat(width);
}
return `${text.slice(0, width - 3)}...`;
}
function padCell(text: string, width: number): string {
const truncated = truncateForWidth(text, width);
return `${truncated}${" ".repeat(Math.max(0, width - truncated.length))}`;
}
function wrapForWidth(text: string, width: number, maxLines: number): string[] {
if (width <= 0 || maxLines <= 0) {
return [];
}
const normalized = text.replace(/\s+/g, " ").trim();
if (!normalized) {
return [];
}
const words = normalized.split(" ");
const lines: string[] = [];
let current = "";
for (const word of words) {
const candidate = current ? `${current} ${word}` : word;
if (candidate.length <= width) {
current = candidate;
continue;
}
if (current) {
lines.push(current);
if (lines.length === maxLines) {
lines[maxLines - 1] = truncateForWidth(lines[maxLines - 1], width);
return lines;
}
}
current = word.length <= width ? word : truncateForWidth(word, width);
}
if (current && lines.length < maxLines) {
lines.push(current);
}
return lines;
}
function getCurrentModelLabel(ctx: ExtensionContext): string {
if (ctx.model) {
return `${ctx.model.provider}/${ctx.model.id}`;
}
const branch = ctx.sessionManager.getBranch();
for (let index = branch.length - 1; index >= 0; index -= 1) {
const entry = branch[index];
if (entry.type === "model_change") {
return `${entry.provider}/${entry.modelId}`;
}
}
return "model not set";
}
function getRecentActivitySummary(ctx: ExtensionContext): string {
const branch = ctx.sessionManager.getBranch();
for (let index = branch.length - 1; index >= 0; index -= 1) {
const entry = branch[index];
if (entry.type !== "message") {
continue;
}
const text = extractMessageText(entry.message).replace(/\s+/g, " ").trim();
if (!text) {
continue;
}
const role = entry.message.role === "assistant"
? "agent"
: entry.message.role === "user"
? "you"
: entry.message.role;
return `${role}: ${text}`;
}
return "No messages yet in this session.";
}
function buildTitledBorder(width: number, title: string): { left: string; right: string } {
const gap = Math.max(0, width - title.length);
const left = Math.floor(gap / 2);
return {
left: "─".repeat(left),
right: "─".repeat(gap - left),
};
}
function formatShortcutLine(command: string, description: string, width: number): string {
const commandWidth = Math.min(18, Math.max(13, Math.floor(width * 0.3)));
return truncateForWidth(`${padCell(command, commandWidth)} ${description}`, width);
}
async function pathExists(path: string): Promise<boolean> {
try {
await stat(path);
return true;
} catch {
return false;
}
}
function buildProjectAgentsTemplate(): string {
return `# Feynman Project Guide
This file is read automatically at startup. It is the durable project memory for Feynman.
## Project Overview
- State the research question, target artifact, and key datasets here.
## Ground Rules
- Do not modify raw data in \`Data/Raw/\` or equivalent raw-data folders.
- Read first, act second: inspect project structure and existing notes before making changes.
- Prefer durable artifacts in \`notes/\`, \`outputs/\`, \`experiments/\`, and \`papers/\`.
- Keep strong claims source-grounded. Include direct URLs in final writeups.
## Current Status
- Replace this section with the latest project status, known issues, and next steps.
## Session Logging
- Use \`/log\` at the end of meaningful sessions to write a durable session note into \`notes/session-logs/\`.
`;
}
function buildSessionLogsReadme(): string {
return `# Session Logs
Use \`/log\` to write one durable note per meaningful Feynman session.
Recommended contents:
- what was done
- strongest findings
- artifacts written
- unresolved questions
- next steps
`;
}
export default function researchTools(pi: ExtensionAPI): void {
function installFeynmanHeader(ctx: ExtensionContext): void {
if (!ctx.hasUI) {
@@ -421,13 +599,101 @@ export default function researchTools(pi: ExtensionAPI): void {
}
ctx.ui.setHeader((_tui, theme) => ({
render(_width: number): string[] {
return [
"",
`${theme.fg("accent", theme.bold("Feynman"))}${theme.fg("muted", " research agent")}`,
theme.fg("dim", "sources first • memory on • scheduled research ready"),
"",
render(width: number): string[] {
const maxAvailableWidth = Math.max(width - 2, 1);
const preferredWidth = Math.min(104, Math.max(56, width - 4));
const cardWidth = Math.min(maxAvailableWidth, preferredWidth);
const innerWidth = cardWidth - 2;
const outerPadding = " ".repeat(Math.max(0, Math.floor((width - cardWidth) / 2)));
const title = truncateForWidth(` Feynman v${FEYNMAN_VERSION} `, innerWidth);
const titledBorder = buildTitledBorder(innerWidth, title);
const modelLabel = getCurrentModelLabel(ctx);
const sessionLabel = ctx.sessionManager.getSessionName()?.trim() || "default session";
const directoryLabel = formatHeaderPath(ctx.cwd);
const recentActivity = getRecentActivitySummary(ctx);
const shortcuts = [
["/lit", "survey papers on a topic"],
["/deepresearch", "run a source-heavy research pass"],
["/draft", "draft a paper-style writeup"],
["/jobs", "inspect active background work"],
];
const lines: string[] = [];
const push = (line: string): void => {
lines.push(`${outerPadding}${line}`);
};
const renderBoxLine = (content: string): string =>
`${theme.fg("borderMuted", "│")}${content}${theme.fg("borderMuted", "│")}`;
const renderDivider = (): string =>
`${theme.fg("borderMuted", "├")}${theme.fg("borderMuted", "─".repeat(innerWidth))}${theme.fg("borderMuted", "┤")}`;
const styleAccentCell = (text: string, cellWidth: number): string =>
theme.fg("accent", theme.bold(padCell(text, cellWidth)));
const styleMutedCell = (text: string, cellWidth: number): string =>
theme.fg("muted", padCell(text, cellWidth));
push("");
push(
theme.fg("borderMuted", `${titledBorder.left}`) +
theme.fg("accent", theme.bold(title)) +
theme.fg("borderMuted", `${titledBorder.right}`),
);
if (innerWidth < 88) {
const activityLines = wrapForWidth(recentActivity, innerWidth, 2);
push(renderBoxLine(padCell("", innerWidth)));
push(renderBoxLine(theme.fg("accent", theme.bold(padCell("Research session ready", innerWidth)))));
push(renderBoxLine(padCell(`model: ${modelLabel}`, innerWidth)));
push(renderBoxLine(padCell(`session: ${sessionLabel}`, innerWidth)));
push(renderBoxLine(padCell(`directory: ${directoryLabel}`, innerWidth)));
push(renderDivider());
push(renderBoxLine(theme.fg("accent", theme.bold(padCell("Quick starts", innerWidth)))));
for (const [command, description] of shortcuts) {
push(renderBoxLine(padCell(formatShortcutLine(command, description, innerWidth), innerWidth)));
}
push(renderDivider());
push(renderBoxLine(theme.fg("accent", theme.bold(padCell("Recent activity", innerWidth)))));
for (const activityLine of activityLines.length > 0 ? activityLines : ["No messages yet in this session."]) {
push(renderBoxLine(padCell(activityLine, innerWidth)));
}
} else {
const leftWidth = Math.min(44, Math.max(38, Math.floor(innerWidth * 0.43)));
const rightWidth = innerWidth - leftWidth - 3;
const activityLines = wrapForWidth(recentActivity, innerWidth, 2);
const row = (
left: string,
right: string,
options?: { leftAccent?: boolean; rightAccent?: boolean; leftMuted?: boolean; rightMuted?: boolean },
): string => {
const leftCell = options?.leftAccent
? styleAccentCell(left, leftWidth)
: options?.leftMuted
? styleMutedCell(left, leftWidth)
: padCell(left, leftWidth);
const rightCell = options?.rightAccent
? styleAccentCell(right, rightWidth)
: options?.rightMuted
? styleMutedCell(right, rightWidth)
: padCell(right, rightWidth);
return renderBoxLine(`${leftCell}${theme.fg("borderMuted", " │ ")}${rightCell}`);
};
push(renderBoxLine(padCell("", innerWidth)));
push(row("Research session ready", "Quick starts", { leftAccent: true, rightAccent: true }));
push(row(`model: ${modelLabel}`, formatShortcutLine(shortcuts[0][0], shortcuts[0][1], rightWidth)));
push(row(`session: ${sessionLabel}`, formatShortcutLine(shortcuts[1][0], shortcuts[1][1], rightWidth)));
push(row(`directory: ${directoryLabel}`, formatShortcutLine(shortcuts[2][0], shortcuts[2][1], rightWidth)));
push(row("ask naturally; slash commands are optional", formatShortcutLine(shortcuts[3][0], shortcuts[3][1], rightWidth), { leftMuted: true }));
push(renderDivider());
push(renderBoxLine(theme.fg("accent", theme.bold(padCell("Recent activity", innerWidth)))));
for (const activityLine of activityLines.length > 0 ? activityLines : ["No messages yet in this session."]) {
push(renderBoxLine(padCell(activityLine, innerWidth)));
}
}
push(theme.fg("borderMuted", `${"─".repeat(innerWidth)}`));
push("");
return lines;
},
invalidate() {},
}));
@@ -441,6 +707,75 @@ export default function researchTools(pi: ExtensionAPI): void {
installFeynmanHeader(ctx);
});
pi.registerCommand("alpha-login", {
description: "Sign in to alphaXiv from inside Feynman.",
handler: async (_args, ctx) => {
if (isAlphaLoggedIn()) {
const name = getAlphaUserName();
ctx.ui.notify(name ? `alphaXiv already connected as ${name}` : "alphaXiv already connected", "info");
return;
}
await loginAlpha();
const name = getAlphaUserName();
ctx.ui.notify(name ? `alphaXiv connected as ${name}` : "alphaXiv login complete", "info");
},
});
pi.registerCommand("alpha-logout", {
description: "Clear alphaXiv auth from inside Feynman.",
handler: async (_args, ctx) => {
logoutAlpha();
ctx.ui.notify("alphaXiv auth cleared", "info");
},
});
pi.registerCommand("alpha-status", {
description: "Show alphaXiv authentication status.",
handler: async (_args, ctx) => {
if (!isAlphaLoggedIn()) {
ctx.ui.notify("alphaXiv not connected", "warning");
return;
}
const name = getAlphaUserName();
ctx.ui.notify(name ? `alphaXiv connected as ${name}` : "alphaXiv connected", "info");
},
});
pi.registerCommand("init", {
description: "Initialize AGENTS.md and session-log folders for a research project.",
handler: async (_args, ctx) => {
const agentsPath = resolvePath(ctx.cwd, "AGENTS.md");
const notesDir = resolvePath(ctx.cwd, "notes");
const sessionLogsDir = resolvePath(notesDir, "session-logs");
const sessionLogsReadmePath = resolvePath(sessionLogsDir, "README.md");
const created: string[] = [];
const skipped: string[] = [];
await mkdir(notesDir, { recursive: true });
await mkdir(sessionLogsDir, { recursive: true });
if (!(await pathExists(agentsPath))) {
await writeFile(agentsPath, buildProjectAgentsTemplate(), "utf8");
created.push("AGENTS.md");
} else {
skipped.push("AGENTS.md");
}
if (!(await pathExists(sessionLogsReadmePath))) {
await writeFile(sessionLogsReadmePath, buildSessionLogsReadme(), "utf8");
created.push("notes/session-logs/README.md");
} else {
skipped.push("notes/session-logs/README.md");
}
const createdSummary = created.length > 0 ? `created: ${created.join(", ")}` : "created: nothing";
const skippedSummary = skipped.length > 0 ? `; kept existing: ${skipped.join(", ")}` : "";
ctx.ui.notify(`${createdSummary}${skippedSummary}`, "info");
},
});
pi.registerTool({
name: "session_search",
label: "Session Search",

View File

@@ -9,6 +9,7 @@
"files": [
"bin/",
"dist/",
".pi/agents/",
".pi/settings.json",
".pi/themes/",
"extensions/",

View File

@@ -4,6 +4,7 @@ description: Compare a paper's claims against its public codebase and identify m
Audit the paper and codebase for: $@
Requirements:
- Prefer the `researcher` subagent for evidence gathering and the `verifier` subagent for the mismatch pass when the audit is non-trivial.
- Identify the canonical paper first with `alpha_search` and `alpha_get_paper`.
- Extract implementation-sensitive claims with `alpha_ask_paper`.
- If a public repo exists, inspect it with `alpha_read_code`.

View File

@@ -4,6 +4,8 @@ description: Turn a research idea into a paper-oriented end-to-end run with lite
Run an autoresearch workflow for: $@
Requirements:
- Prefer the project `auto` chain or the `planner` + `researcher` + `verifier` + `writer` subagents when the task is broad enough to benefit from decomposition.
- If the run is likely to take a while, or the user wants it detached, launch the subagent workflow in background with `clarify: false, async: true` and report how to inspect status.
- Start by clarifying the research objective, scope, and target contribution.
- Search for the strongest relevant primary sources first.
- If the topic is current, product-oriented, market-facing, or asks about latest developments, start with `web_search` and `fetch_content`.

View File

@@ -4,6 +4,7 @@ description: Compare multiple sources on a topic and produce a source-grounded m
Compare sources for: $@
Requirements:
- Use the `researcher` subagent to gather source material when the comparison set is broad, and the `verifier` subagent to pressure-test the resulting matrix when needed.
- Identify the strongest relevant primary sources first.
- For current or market-facing topics, use `web_search` and `fetch_content` to gather up-to-date primary sources before comparing them.
- For academic claims, use `alpha_search` and inspect the strongest papers directly.

View File

@@ -4,6 +4,8 @@ description: Run a thorough, source-heavy investigation on a topic and produce a
Run a deep research workflow for: $@
Requirements:
- If the task is broad, multi-source, or obviously long-running, prefer delegating through the `subagent` tool. Use the project `researcher`, `verifier`, and `writer` agents, or the project `deep` chain when that decomposition fits.
- If the user wants it to run unattended, or the sweep will clearly take a while, prefer background execution with `subagent` using `clarify: false, async: true`, then report how to inspect status.
- If the topic is current, product-oriented, market-facing, regulatory, or asks about latest developments, start with `web_search` and `fetch_content`.
- If the topic has an academic literature component, use `alpha_search`, `alpha_get_paper`, and `alpha_ask_paper` for the strongest papers.
- Do not rely on a single source type when the topic spans both current reality and academic background.

View File

@@ -4,6 +4,7 @@ description: Turn research findings into a polished paper-style draft with equat
Write a paper-style draft for: $@
Requirements:
- Prefer the `writer` subagent when the draft should be produced from already-collected notes, and use `verifier` first if the evidence still looks shaky.
- Ground every claim in inspected sources, experiments, or explicit inference.
- Use clean Markdown structure with LaTeX where equations materially help.
- Include at minimum:

14
prompts/jobs.md Normal file
View File

@@ -0,0 +1,14 @@
---
description: Inspect active background research work, including running processes and scheduled follow-ups.
---
Inspect active background work for this project.
Requirements:
- Use the `process` tool with the `list` action to inspect running and finished managed background processes.
- Use the scheduling tooling to list active recurring or deferred jobs if any are configured.
- Summarize:
- active background processes
- queued or recurring research watches
- failures that need attention
- the next concrete command the user should run if they want logs or detailed status
- Be concise and operational.

View File

@@ -4,6 +4,7 @@ description: Run a literature review on a topic using paper search and primary-s
Investigate the following topic as a literature review: $@
Requirements:
- Use the `researcher` subagent when the sweep is wide enough to benefit from delegated paper triage before synthesis.
- If the topic is academic or paper-centric, use `alpha_search` first.
- If the topic is current, product-oriented, market-facing, or asks about latest developments, use `web_search` and `fetch_content` first, then use `alpha_search` only for academic background.
- Use `alpha_get_paper` on the most relevant papers before making strong claims.

12
prompts/log.md Normal file
View File

@@ -0,0 +1,12 @@
---
description: Write a durable session log with completed work, findings, open questions, and next steps.
---
Write a session log for the current research work.
Requirements:
- Summarize what was done in this session.
- Capture the strongest findings or decisions.
- List open questions, unresolved risks, and concrete next steps.
- Reference any important artifacts written to `notes/`, `outputs/`, `experiments/`, or `papers/`.
- If any external claims matter, include direct source URLs.
- Save the log to `notes/` as markdown with a date-oriented filename.

View File

@@ -4,6 +4,7 @@ description: Produce a general research memo grounded in explicit sources and di
Write a research memo about: $@
Requirements:
- Use the `researcher` and `writer` subagents when decomposition will improve quality or reduce context pressure.
- Start by finding the strongest relevant sources.
- If the topic is current, market-facing, product-oriented, regulatory, or asks about latest developments, use `web_search` and `fetch_content` first.
- Use `alpha_search` for academic background where relevant, but do not rely on it alone for current topics.

View File

@@ -4,6 +4,7 @@ description: Build a prioritized reading list on a research topic with rationale
Create a research reading list for: $@
Requirements:
- Use the `researcher` subagent when a wider literature sweep would help before curating the final list.
- If the topic is academic, use `alpha_search` with `all` mode.
- If the topic is current, product-oriented, or asks for the latest landscape, use `web_search` and `fetch_content` first, then add `alpha_search` for academic background when relevant.
- Inspect the strongest papers or primary sources directly before recommending them.

View File

@@ -4,6 +4,7 @@ description: Plan or execute a replication workflow for a paper, claim, or bench
Design a replication plan for: $@
Requirements:
- Use the `subagent` tool for decomposition when the replication needs separate planning, evidence extraction, and execution passes.
- Identify the canonical paper or source material first.
- Use `alpha_get_paper` for the target paper.
- Use `alpha_ask_paper` to extract the exact implementation or evaluation details you still need.

14
prompts/watch.md Normal file
View File

@@ -0,0 +1,14 @@
---
description: Set up a recurring or deferred research watch on a topic, company, paper area, or product surface.
---
Create a research watch for: $@
Requirements:
- Start with a baseline sweep of the topic using the strongest relevant sources.
- If the watch is about current events, products, markets, regulations, or releases, use `web_search` and `fetch_content` first.
- If the watch has a literature component, add `alpha_search` and inspect the strongest papers directly.
- Summarize what should be monitored, what signals matter, and what counts as a meaningful change.
- Use `schedule_prompt` to create the recurring or delayed follow-up instead of merely promising to check later.
- If the user wants detached execution for the initial sweep, use `subagent` in background mode and report how to inspect status.
- Save a durable baseline artifact to `outputs/`.
- End with a `Sources` section containing direct URLs for every source used.

View File

@@ -9,6 +9,7 @@ const piPackageRoot = resolve(appRoot, "node_modules", "@mariozechner", "pi-codi
const packageJsonPath = resolve(piPackageRoot, "package.json");
const cliPath = resolve(piPackageRoot, "dist", "cli.js");
const interactiveModePath = resolve(piPackageRoot, "dist", "modes", "interactive", "interactive-mode.js");
const footerPath = resolve(piPackageRoot, "dist", "modes", "interactive", "components", "footer.js");
const workspaceRoot = resolve(appRoot, ".pi", "npm", "node_modules");
const webAccessPath = resolve(workspaceRoot, "pi-web-access", "index.ts");
const sessionSearchIndexerPath = resolve(
@@ -102,6 +103,42 @@ if (existsSync(interactiveModePath)) {
}
}
if (existsSync(footerPath)) {
const footerSource = readFileSync(footerPath, "utf8");
const footerOriginal = [
' // Add thinking level indicator if model supports reasoning',
' let rightSideWithoutProvider = modelName;',
' if (state.model?.reasoning) {',
' const thinkingLevel = state.thinkingLevel || "off";',
' rightSideWithoutProvider =',
' thinkingLevel === "off" ? `${modelName} • thinking off` : `${modelName} • ${thinkingLevel}`;',
' }',
' // Prepend the provider in parentheses if there are multiple providers and there\'s enough room',
' let rightSide = rightSideWithoutProvider;',
' if (this.footerData.getAvailableProviderCount() > 1 && state.model) {',
' rightSide = `(${state.model.provider}) ${rightSideWithoutProvider}`;',
].join("\n");
const footerReplacement = [
' // Add thinking level indicator if model supports reasoning',
' const modelLabel = theme.fg("accent", modelName);',
' let rightSideWithoutProvider = modelLabel;',
' if (state.model?.reasoning) {',
' const thinkingLevel = state.thinkingLevel || "off";',
' const separator = theme.fg("dim", " • ");',
' rightSideWithoutProvider = thinkingLevel === "off"',
' ? `${modelLabel}${separator}${theme.fg("muted", "thinking off")}`',
' : `${modelLabel}${separator}${theme.getThinkingBorderColor(thinkingLevel)(thinkingLevel)}`;',
' }',
' // Prepend the provider in parentheses if there are multiple providers and there\'s enough room',
' let rightSide = rightSideWithoutProvider;',
' if (this.footerData.getAvailableProviderCount() > 1 && state.model) {',
' rightSide = `${theme.fg("muted", `(${state.model.provider})`)} ${rightSideWithoutProvider}`;',
].join("\n");
if (footerSource.includes(footerOriginal)) {
writeFileSync(footerPath, footerSource.replace(footerOriginal, footerReplacement), "utf8");
}
}
if (existsSync(webAccessPath)) {
const source = readFileSync(webAccessPath, "utf8");
if (source.includes('pi.registerCommand("search",')) {

View File

@@ -16,6 +16,8 @@ Operating rules:
- Never answer a latest/current question from arXiv or alpha-backed paper search alone.
- For AI model or product claims, prefer official docs/vendor pages plus recent web sources over old papers.
- Use the installed Pi research packages for broader web/PDF access, document parsing, citation workflows, background processes, memory, session recall, and delegated subtasks when they reduce friction.
- Feynman ships project subagents for research work. Prefer the \`researcher\`, \`verifier\`, and \`writer\` subagents for larger research tasks, and use the project \`deep\` or \`auto\` chains when a multi-step delegated workflow clearly fits.
- 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\`.
- 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.
@@ -23,6 +25,7 @@ Operating rules:
- Feynman is intended to support always-on research work. Use the scheduling package when recurring or deferred work is appropriate instead of telling the user to remember manually.
- Use \`schedule_prompt\` for recurring scans, delayed follow-ups, reminders, and periodic research jobs.
- If the user asks you to remind, check later, run something nightly, or keep watching something over time, call \`schedule_prompt\`. Do not just promise to do it later.
- 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.
- Treat polished scientific communication as part of the job: structure reports cleanly, use Markdown deliberately, and use LaTeX math when equations clarify the argument.

View File

@@ -1,7 +1,7 @@
import "dotenv/config";
import { spawn, spawnSync } from "node:child_process";
import { existsSync, mkdirSync, readFileSync, writeFileSync } from "node:fs";
import { existsSync, mkdirSync, readdirSync, readFileSync, writeFileSync } from "node:fs";
import { homedir } from "node:os";
import { dirname, resolve } from "node:path";
import { stdin as input, stdout as output } from "node:process";
@@ -23,24 +23,214 @@ import {
import { buildFeynmanSystemPrompt } from "./feynman-prompt.js";
type ThinkingLevel = "off" | "low" | "medium" | "high";
type Rgb = { r: number; g: number; b: number };
type ThemeColorValue = string | number;
type ThemeJson = {
$schema?: string;
name: string;
vars?: Record<string, ThemeColorValue>;
colors: Record<string, ThemeColorValue>;
export?: Record<string, ThemeColorValue>;
};
const OSC11_QUERY = "\u001b]11;?\u0007";
const OSC11_RESPONSE_PATTERN =
/\u001b]11;(?:rgb:([0-9a-fA-F]{2,4})\/([0-9a-fA-F]{2,4})\/([0-9a-fA-F]{2,4})|#?([0-9a-fA-F]{6}))(?:\u0007|\u001b\\)/;
const DEFAULT_SAGE_RGB: Rgb = { r: 0x88, g: 0xa8, b: 0x8a };
function parseHexComponent(component: string): number {
const value = Number.parseInt(component, 16);
if (Number.isNaN(value)) {
throw new Error(`Invalid OSC 11 component: ${component}`);
}
if (component.length === 2) {
return value;
}
return Math.round(value / ((1 << (component.length * 4)) - 1) * 255);
}
function parseHexColor(color: string): Rgb | undefined {
const match = color.trim().match(/^#?([0-9a-fA-F]{6})$/);
if (!match) {
return undefined;
}
return {
r: Number.parseInt(match[1].slice(0, 2), 16),
g: Number.parseInt(match[1].slice(2, 4), 16),
b: Number.parseInt(match[1].slice(4, 6), 16),
};
}
function rgbToHex(rgb: Rgb): string {
return `#${[rgb.r, rgb.g, rgb.b]
.map((value) => Math.max(0, Math.min(255, Math.round(value))).toString(16).padStart(2, "0"))
.join("")}`;
}
function blendRgb(base: Rgb, tint: Rgb, alpha: number): Rgb {
const mix = (baseChannel: number, tintChannel: number) =>
baseChannel + (tintChannel - baseChannel) * alpha;
return {
r: mix(base.r, tint.r),
g: mix(base.g, tint.g),
b: mix(base.b, tint.b),
};
}
function isLightRgb(rgb: Rgb): boolean {
const luminance = (0.299 * rgb.r + 0.587 * rgb.g + 0.114 * rgb.b) / 255;
return luminance >= 0.6;
}
function resolveThemeColorValue(
value: ThemeColorValue | undefined,
vars: Record<string, ThemeColorValue> | undefined,
visited = new Set<string>(),
): ThemeColorValue | undefined {
if (value === undefined || typeof value === "number" || value === "" || value.startsWith("#")) {
return value;
}
if (!vars || !(value in vars) || visited.has(value)) {
return value;
}
visited.add(value);
return resolveThemeColorValue(vars[value], vars, visited);
}
function resolveThemeRgb(
value: ThemeColorValue | undefined,
vars: Record<string, ThemeColorValue> | undefined,
): Rgb | undefined {
const resolved = resolveThemeColorValue(value, vars);
return typeof resolved === "string" ? parseHexColor(resolved) : undefined;
}
function deriveMessageBackgrounds(themeJson: ThemeJson, terminalBackgroundHex: string): Pick<ThemeJson["colors"], "userMessageBg" | "customMessageBg"> | undefined {
const terminalBackground = parseHexColor(terminalBackgroundHex);
if (!terminalBackground) {
return undefined;
}
const tint =
resolveThemeRgb(themeJson.colors.accent, themeJson.vars) ??
resolveThemeRgb(themeJson.vars?.sage, themeJson.vars) ??
DEFAULT_SAGE_RGB;
const lightBackground = isLightRgb(terminalBackground);
const userAlpha = lightBackground ? 0.15 : 0.23;
const customAlpha = lightBackground ? 0.11 : 0.17;
return {
userMessageBg: rgbToHex(blendRgb(terminalBackground, tint, userAlpha)),
customMessageBg: rgbToHex(blendRgb(terminalBackground, tint, customAlpha)),
};
}
async function probeTerminalBackgroundHex(timeoutMs = 120): Promise<string | undefined> {
if (typeof process.env.FEYNMAN_TERMINAL_BG === "string" && process.env.FEYNMAN_TERMINAL_BG.trim()) {
return process.env.FEYNMAN_TERMINAL_BG.trim();
}
if (typeof process.env.PI_TERMINAL_BG === "string" && process.env.PI_TERMINAL_BG.trim()) {
return process.env.PI_TERMINAL_BG.trim();
}
if (!input.isTTY || !output.isTTY || typeof input.setRawMode !== "function") {
return undefined;
}
const wasRaw = "isRaw" in input ? Boolean((input as typeof input & { isRaw?: boolean }).isRaw) : false;
return await new Promise<string | undefined>((resolve) => {
let settled = false;
let buffer = "";
const finish = (value: string | undefined) => {
if (settled) {
return;
}
settled = true;
clearTimeout(timer);
input.off("data", onData);
try {
if (!wasRaw) {
input.setRawMode(false);
}
} catch {
// Ignore raw mode restore failures and return best-effort detection.
}
resolve(value);
};
const onData = (chunk: string | Buffer) => {
buffer += chunk.toString("utf8");
const match = buffer.match(OSC11_RESPONSE_PATTERN);
if (!match) {
if (buffer.length > 512) {
finish(undefined);
}
return;
}
if (match[4]) {
finish(`#${match[4].toLowerCase()}`);
return;
}
try {
finish(
rgbToHex({
r: parseHexComponent(match[1]),
g: parseHexComponent(match[2]),
b: parseHexComponent(match[3]),
}),
);
} catch {
finish(undefined);
}
};
const timer = setTimeout(() => finish(undefined), timeoutMs);
input.on("data", onData);
try {
if (!wasRaw) {
input.setRawMode(true);
}
output.write(OSC11_QUERY);
} catch {
finish(undefined);
}
});
}
function printHelp(): void {
console.log(`Feynman commands:
/help Show this help
/init Initialize AGENTS.md and session-log folders
/alpha-login Sign in to alphaXiv
/alpha-logout Clear alphaXiv auth
/alpha-status Show alphaXiv auth status
/new Start a fresh persisted session
/exit Quit the REPL
/lit-review <topic> Expand the literature review prompt template
/lit <topic> Expand the literature review prompt template
/replicate <paper> Expand the replication prompt template
/reading-list <topic> Expand the reading list prompt template
/research-memo <topic> Expand the general research memo prompt template
/reading <topic> Expand the reading list prompt template
/memo <topic> Expand the general research memo prompt template
/deepresearch <topic> Expand the thorough source-heavy research prompt template
/autoresearch <idea> Expand the idea-to-paper autoresearch prompt template
/compare-sources <topic> Expand the source comparison prompt template
/paper-code-audit <item> Expand the paper/code audit prompt template
/paper-draft <topic> Expand the paper-style writing prompt template
/compare <topic> Expand the source comparison prompt template
/audit <item> Expand the paper/code audit prompt template
/draft <topic> Expand the paper-style writing prompt template
/log Write a durable session log
/watch <topic> Create a recurring or deferred research watch
/jobs Inspect active background work
Package-powered workflows:
/agents Open the subagent and chain manager
/run /chain /parallel Delegate research work to subagents
/ps Open the background process panel
/schedule-prompt Manage deferred and recurring jobs
/search Search prior indexed sessions
/preview Preview generated markdown or code artifacts
CLI flags:
--prompt "<text>" Run one prompt and exit
@@ -116,6 +306,7 @@ function patchEmbeddedPiBranding(piPackageRoot: string): void {
const packageJsonPath = resolve(piPackageRoot, "package.json");
const cliPath = resolve(piPackageRoot, "dist", "cli.js");
const interactiveModePath = resolve(piPackageRoot, "dist", "modes", "interactive", "interactive-mode.js");
const footerPath = resolve(piPackageRoot, "dist", "modes", "interactive", "components", "footer.js");
if (existsSync(packageJsonPath)) {
const pkg = JSON.parse(readFileSync(packageJsonPath, "utf8")) as {
@@ -149,6 +340,42 @@ function patchEmbeddedPiBranding(piPackageRoot: string): void {
);
}
}
if (existsSync(footerPath)) {
const footerSource = readFileSync(footerPath, "utf8");
const footerOriginal = [
' // Add thinking level indicator if model supports reasoning',
' let rightSideWithoutProvider = modelName;',
' if (state.model?.reasoning) {',
' const thinkingLevel = state.thinkingLevel || "off";',
' rightSideWithoutProvider =',
' thinkingLevel === "off" ? `${modelName} • thinking off` : `${modelName} • ${thinkingLevel}`;',
' }',
' // Prepend the provider in parentheses if there are multiple providers and there\'s enough room',
' let rightSide = rightSideWithoutProvider;',
' if (this.footerData.getAvailableProviderCount() > 1 && state.model) {',
' rightSide = `(${state.model.provider}) ${rightSideWithoutProvider}`;',
].join("\n");
const footerReplacement = [
' // Add thinking level indicator if model supports reasoning',
' const modelLabel = theme.fg("accent", modelName);',
' let rightSideWithoutProvider = modelLabel;',
' if (state.model?.reasoning) {',
' const thinkingLevel = state.thinkingLevel || "off";',
' const separator = theme.fg("dim", " • ");',
' rightSideWithoutProvider = thinkingLevel === "off"',
' ? `${modelLabel}${separator}${theme.fg("muted", "thinking off")}`',
' : `${modelLabel}${separator}${theme.getThinkingBorderColor(thinkingLevel)(thinkingLevel)}`;',
' }',
' // Prepend the provider in parentheses if there are multiple providers and there\'s enough room',
' let rightSide = rightSideWithoutProvider;',
' if (this.footerData.getAvailableProviderCount() > 1 && state.model) {',
' rightSide = `${theme.fg("muted", `(${state.model.provider})`)} ${rightSideWithoutProvider}`;',
].join("\n");
if (footerSource.includes(footerOriginal)) {
writeFileSync(footerPath, footerSource.replace(footerOriginal, footerReplacement), "utf8");
}
}
}
function patchPackageWorkspace(appRoot: string): void {
@@ -491,7 +718,28 @@ function setupPreviewDependencies(): void {
throw new Error("Automatic preview setup is only supported on macOS with Homebrew right now.");
}
function syncFeynmanTheme(appRoot: string, agentDir: string): void {
function syncDirectory(sourceDir: string, targetDir: string): void {
if (!existsSync(sourceDir)) {
return;
}
mkdirSync(targetDir, { recursive: true });
for (const entry of readdirSync(sourceDir, { withFileTypes: true })) {
const sourcePath = resolve(sourceDir, entry.name);
const targetPath = resolve(targetDir, entry.name);
if (entry.isDirectory()) {
syncDirectory(sourcePath, targetPath);
continue;
}
if (entry.isFile()) {
writeFileSync(targetPath, readFileSync(sourcePath, "utf8"), "utf8");
}
}
}
function syncFeynmanTheme(appRoot: string, agentDir: string, terminalBackgroundHex?: string): void {
const sourceThemePath = resolve(appRoot, ".pi", "themes", "feynman.json");
const targetThemeDir = resolve(agentDir, "themes");
const targetThemePath = resolve(targetThemeDir, "feynman.json");
@@ -501,7 +749,36 @@ function syncFeynmanTheme(appRoot: string, agentDir: string): void {
}
mkdirSync(targetThemeDir, { recursive: true });
writeFileSync(targetThemePath, readFileSync(sourceThemePath, "utf8"), "utf8");
const sourceTheme = readFileSync(sourceThemePath, "utf8");
if (!terminalBackgroundHex) {
writeFileSync(targetThemePath, sourceTheme, "utf8");
return;
}
try {
const parsedTheme = JSON.parse(sourceTheme) as ThemeJson;
const derivedBackgrounds = deriveMessageBackgrounds(parsedTheme, terminalBackgroundHex);
if (!derivedBackgrounds) {
writeFileSync(targetThemePath, sourceTheme, "utf8");
return;
}
const generatedTheme: ThemeJson = {
...parsedTheme,
colors: {
...parsedTheme.colors,
...derivedBackgrounds,
},
};
writeFileSync(targetThemePath, JSON.stringify(generatedTheme, null, 2) + "\n", "utf8");
} catch {
writeFileSync(targetThemePath, sourceTheme, "utf8");
}
}
function syncFeynmanAgents(appRoot: string, agentDir: string): void {
syncDirectory(resolve(appRoot, ".pi", "agents"), resolve(agentDir, "agents"));
}
async function main(): Promise<void> {
@@ -539,9 +816,11 @@ async function main(): Promise<void> {
const workingDir = resolve(values.cwd ?? process.cwd());
const sessionDir = resolve(values["session-dir"] ?? resolve(homedir(), ".feynman", "sessions"));
const terminalBackgroundHex = await probeTerminalBackgroundHex();
mkdirSync(sessionDir, { recursive: true });
mkdirSync(feynmanAgentDir, { recursive: true });
syncFeynmanTheme(appRoot, feynmanAgentDir);
syncFeynmanTheme(appRoot, feynmanAgentDir, terminalBackgroundHex);
syncFeynmanAgents(appRoot, feynmanAgentDir);
const feynmanSettingsPath = resolve(feynmanAgentDir, "settings.json");
const feynmanAuthPath = resolve(feynmanAgentDir, "auth.json");
const thinkingLevel = normalizeThinkingLevel(values.thinking ?? process.env.FEYNMAN_THINKING) ?? "medium";
@@ -637,6 +916,8 @@ async function main(): Promise<void> {
...process.env,
PI_CODING_AGENT_DIR: feynmanAgentDir,
FEYNMAN_CODING_AGENT_DIR: feynmanAgentDir,
FEYNMAN_TERMINAL_BG: terminalBackgroundHex,
PI_TERMINAL_BG: terminalBackgroundHex,
FEYNMAN_PI_NPM_ROOT: resolve(appRoot, ".pi", "npm", "node_modules"),
FEYNMAN_SESSION_DIR: sessionDir,
PI_SESSION_DIR: sessionDir,