claude-config
AGENTS.md1 starsChanged 4 months agoMITView on GitHub
Nine subagents in agents/. Each is a separate context with its own system prompt, model, and tool set. The table in .claude/rules/common/agents.md is the authoritative roster: the review rules forbid invoking a reviewer agent that is not listed there, so an agent file without a roster row is dead weight. /add-agent keeps the two in sync; /audit-config reports when they drift. Least privilege. Reviewers get Read, Grep, Glob, Bash and nothing more. Giving a reviewer Write means it will quietly…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 31,774 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository14 other files it gives its agents.
- .claude/rules/common/agents.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- standards-<layer>Skill
- add-agentSkill
- add-standards-layerSkill
- audit-configSkill
- plan-featureSkill
- research-reuseSkill
- review-changesSkill
- shipSkill
- standards-fastapiSkill
- standards-pythonSkill
- standards-typescriptSkill
- tddSkill
The file
We split this agents.md into the rules it gives an agent
1 rule. About 0 are common, 0 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈0
Distinctive · estimate
This author's own choices. Worth a look.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 43
Run /add-agent <name>, or follow extending.md manually.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 AGENTS.md files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Agents
Nine subagents in agents/. Each is a separate context with its own system prompt,
model, and tool set.
The table in .claude/rules/common/agents.md is the
authoritative roster: the review rules forbid invoking a reviewer agent that is not listed there,
so an agent file without a roster row is dead weight. /add-agent keeps the two in sync;
/audit-config reports when they drift.
| Agent | Model | Tools | Use for |
planner | opus | read-only | Phased implementation plans for complex features and refactors |
architect | opus | read-only | System design: module boundaries, data flow, technology trade-offs |
tdd-guide | sonnet | read + write | Driving features and bug fixes through RED → GREEN → refactor |
code-reviewer | sonnet | read-only | Quality, patterns, and best practices after code is written |
security-reviewer | opus | read-only | OWASP-focused audit before commits and on sensitive changes |
build-error-resolver | sonnet | read + write | Getting a failing build, compile, or type-check back to green |
e2e-runner | sonnet | read + bash | Running and triaging end-to-end tests for critical flows |
refactor-cleaner | sonnet | read + write | Dead-code removal and structural cleanup without behaviour change |
doc-updater | haiku | read + write | Keeping README, docstrings, comments, and changelogs in sync |
Design rules
Least privilege. Reviewers get Read, Grep, Glob, Bash and nothing more. Giving a reviewer
Write means it will quietly fix what it should have reported, and the finding never reaches you.
Model matched to the cost of being wrong. Opus where a miss is expensive (architecture, security), Sonnet for the bulk of the work, Haiku for mechanical documentation updates. Reserving the strongest model for everything is as wrong as reserving it for nothing.
Parallel by default. Independent reviews run at the same time. /review-changes launches
code-reviewer and security-reviewer together rather than in sequence.
Stack-specific reviewers
Language reviewers (python-reviewer, typescript-reviewer, rust-reviewer) are not shipped by
default. Add one with /add-agent when a project needs it — the skill will register it in the
roster, which is what makes it invocable under the review rules.
Adding an agent
Run /add-agent <name>, or follow extending.md manually.
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.