Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add commands/justinjdev/fellowship/red-bookgit clone --depth 1 https://github.com/justinjdev/fellowshipWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00031 | $0.00811 |
| Opus 5 | $0.00015 | $0.00405 |
| Sonnet 5 | $0.00006 | $0.00162 |
| Haiku 4.5 | $0.00003 | $0.00081 |
Grade A, and why
red-book scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 2d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 80 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Red Book — Learn From PR Reviews
Overview
Extracts conventions from PR review feedback and adds them to CLAUDE.md. This is the missing piece of the convention learning loop: chronicle bootstraps conventions, gather-lore studies them, warden enforces them, and this skill captures new ones from real reviewer feedback.
When to Use
- After receiving PR review comments that reveal a convention you didn't know about
- After a "wrong approach" rejection — the reviewer's feedback is the most valuable signal
- Periodically, to process accumulated review feedback across recent PRs
Process
Step 1: Gather Feedback
Ask the user for the source:
"Paste the review comments, or give me a PR URL and I'll read the comments."
If given a PR URL, use gh api repos/{owner}/{repo}/pulls/{number}/comments and gh api repos/{owner}/{repo}/pulls/{number}/reviews to fetch review comments.
Step 2: Classify Comments
For each review comment, classify it:
| Category | Description | Action |
|---|---|---|
| Convention | "We don't do it that way" — reveals a pattern or rule | Extract as a convention |
| Bug/Logic | Points out a functional error | Skip — not a convention |
| Nit/Style | Minor formatting preference | Extract only if it recurs across reviews |
| Question | Reviewer asking for clarification | Skip — not a convention |
Present the classification to the user: "Here's how I categorized the feedback. Any I got wrong?"
Step 3: Extract Conventions
For each comment classified as Convention, formalize it:
- **[Category]: [Rule]** — [Why, from reviewer's comment]
Reference: [file:line from the PR where this was flagged]
Categories should match existing CLAUDE.md sections (e.g., Structure, Error Handling, Naming, Data Flow). If a comment doesn't fit an existing category, propose a new one.
Step 4: Check for Duplicates
Read the current CLAUDE.md ## Review Conventions section. For each extracted convention:
- If it's already documented: skip, note that it was missed during implementation (warden should have caught it)
- If it's a refinement of an existing rule: propose updating the existing rule
- If it's new: propose adding it
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 2d ago First seen · 80 lines · 31 tokens per session scan A 291d9d604414
red-book is a command published in the GitHub repository justinjdev/fellowship (5 stars, last pushed 19d ago), licensed Apache-2.0. It adds 31 tokens to every session and 811 once invoked, about $0.0002 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other commands, from other repositories
auto-mode
Idea-to-running-code lifecycle orchestration. 10-phase pipeline with 5 hard decision gates, wave-based parallelism, and STATE.json resumability. Composes /deep-research, /auto-swarm-nth, /production-upgrade, /security-audit, and /ship into a single end-to-end flow.
max-research
Nuclear-scale autonomous research — deploys 500-1000 agents in ONE massive simultaneous wave for exhaustive topic saturation. Deep-research methodology × auto-swarm scale = maximum parallel intelligence. WARNING: Extreme resource consumption.
omni-plan
ProductionOS flagship — 13-step orchestrative pipeline with tri-tiered evaluation, recursive convergence, CEO/Eng/Design review chain, CLEAR framework evaluation, multi-model judge tribunal, and autonomous PIVOT/REFINE/PROCEED decisions. Targets 100% production-ready output.
auto-swarm-nth
Nth-iteration agent swarm — spawns parallel agent waves, evaluates strictly per wave, re-swarms gaps until 100% coverage and 10/10 quality. Can invoke any ProductionOS skill or command within waves.
frontend-upgrade
Full-stack frontend upgrade pipeline — fuses /production-upgrade iterative audit with /plan-ceo-review vision and /plan-eng-review rigor. Deploys parallel auto-swarm agents for iterative audit and execution. Enriched with /deep-research for competitive parity.
omni-plan-nth
Nth-iteration omni-plan — recursive orchestration that chains ALL ProductionOS skills and agents, evaluates strictly per iteration, and loops until 10/10 is achieved. Each iteration can invoke any command or skill in the system.