Two new tool-operation skills with deep references/ docs: - opencode-cli: SKILL.md + 6 references covering rules, agents, models, commands, formatters, permissions, skills, MCP, plugins, custom tools, LSP, themes, keybinds, server API, SDK, GitHub Actions, IDE, network, troubleshooting (full opencode.ai/docs surface) - feynman-cli: SKILL.md + 6 references covering install, setup, config, CLI, REPL slash commands, agents/tools/packages, and full pi-subagents custom-agent spec (verified against the working install) Migrate 12 skills from ~/.claude/skills into _shared/community-skills/: - clean copy: intel-briefing, vercel-react-best-practices, ui-ux-pro-max - core-only: notebooklm (data/images stripped — 184M to 224K) - light sanitize: anythingllm-manager (gitea URL), foia-tool (DB password), jira (atlassian instance + email), librarian (paths), obsidian-tasks (vault path + email-in-cred-path) - branding sanitize: marketing-strategist + pentest-reporter (Proudsec variants normalized to <COMPANY>) - secrets sanitize: waha-whatsapp (IP, API key, vault path placeholders) Skipped per user: proudguard-api (kept locally only). build.py: - DEFAULT_SKILL_PERSONA_MAP: 14 new entries - NAME_PATTERNS: opencode + jira to coding-tools; notebooklm + feynman- to ai-llm-dev; waha- to osint-intel Community-skills index: 703 -> 716 (+13). Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
6.2 KiB
CLI Reference
Core commands
| Command | Purpose |
|---|---|
feynman |
Launch interactive REPL in cwd |
feynman chat [prompt] |
Start chat explicitly, optional initial prompt |
feynman --prompt "<text>" |
One-shot prompt: print response, exit |
feynman help |
Show CLI help |
feynman setup |
Guided setup wizard (model, auth, packages) |
feynman setup preview |
Install / verify pandoc for /preview |
feynman doctor |
Diagnose config, auth, Pi runtime, preview deps |
feynman status |
Current setup summary (model + auth + packages) |
feynman --version |
Print version |
Model management
feynman model list # list every reachable model
feynman model login <provider> # OAuth or API key
feynman model logout <provider> # clear stored auth
feynman model set <provider/model-id> # set default; accepts
# provider/model-name
# provider:model-name
Examples:
feynman model set anthropic/claude-sonnet-4-20250514
feynman model set openai:gpt-4o
feynman model set github-copilot/gpt-5.3-codex
AlphaXiv (academic search)
| Command | Purpose |
|---|---|
feynman alpha login |
Sign in to alphaXiv |
feynman alpha logout |
Clear alphaXiv auth |
feynman alpha status |
Check alphaXiv auth status |
Without alphaXiv auth Feynman falls back to general web search and loses
citation metadata, discussion threads, and full-text PDFs. Strongly
recommend running feynman alpha login before serious research.
Packages
| Command | Purpose |
|---|---|
feynman packages list |
All packages + install status |
feynman packages install <preset> |
Install optional package preset |
feynman update |
Update all packages |
feynman update <package> |
Update one package |
Optional preset today: generative-ui (macOS only — interactive HTML
widgets). Other names hidden on Linux/Windows.
Workflow shortcuts (CLI form)
These are equivalent to the REPL slash commands but run as one-shots from the shell:
feynman deepresearch "topic"
feynman lit "topic" # literature review
feynman review draft.md # peer review of an artifact
feynman audit 2401.12345 # paper-vs-code audit (arXiv id)
feynman replicate "claim or paper"
feynman compare "topic across sources"
feynman draft "paper-style draft topic"
CLI form auto-creates a session, runs the workflow, prints/saves the
artifact, and exits. Pair with --cwd to scope research to a directory:
feynman --cwd ~/research/llm-evals lit "agentic LLM benchmarks"
Search
| Command | Purpose |
|---|---|
feynman search status |
Show Pi web-access status |
Global flags
| Flag | Effect |
|---|---|
--prompt "<text>" |
One-shot prompt → exit |
--model <provider/id> |
Override default model for this run |
--thinking <level> |
`off |
--cwd <path> |
Run with a specific working directory |
--session-dir <path> |
Custom session storage path |
--new-session |
Force fresh session |
--alpha-login |
Alias for feynman alpha login |
--alpha-logout |
Alias for feynman alpha logout |
--alpha-status |
Alias for feynman alpha status |
--doctor |
Alias for feynman doctor |
--setup-preview |
Alias for feynman setup preview |
CI / scripted invocation
ANTHROPIC_API_KEY=$KEY \
FEYNMAN_THINKING=low \
feynman --prompt "Summarize $(cat README.md | head -200)" > brief.md
Notes:
- Pure CLI runs print the model output to stdout; pipe + redirect freely.
--thinking lowkeeps cost down for batch jobs.- Set the appropriate
*_API_KEYenv var so you don't depend onauth.jsonbeing warm (env auth bypasses interactive prompts).
Examples by goal
Quickly summarize a paper:
feynman --prompt "Summarize the key findings of Attention Is All You Need"
Run a deep-research and save the brief:
feynman --cwd ~/research deepresearch "agentic memory persistence patterns"
ls ~/research/outputs/
Switch model just for this run:
feynman --model anthropic/claude-opus-4-5 --thinking high
Verify install + doctor in one shot (CI):
feynman --version && feynman doctor
Commands not present (yet)
The docs claim some, the CLI lacks others. Verify with feynman help if a
command isn't in the reference above. Missing as of this skill snapshot:
feynman run, feynman version (use --version), feynman uninstall,
feynman session, feynman auth, feynman config. Use the alternates:
- session control → REPL
/sessionsor--session-dir - auth →
feynman model login/logoutandfeynman alpha login/logout - config → edit
~/.feynman/agent/settings.jsondirectly