claude-toolkit
CLAUDE.md4 starsChanged 3 months agoMITView on GitHub
Team path (/team-execute) — >=2 source files or design decisions: - Principal Engineer: architecture, language/platform conventions, security review - QA Engineer: test quality, mutation testing, integration verification, PR readiness - Workers: implementation in isolated worktrees (TDD enforced by tdd-guard.sh hook on all Write/Edit, both team and solo paths) - See agents/ for role definitions. See /team-execute for orchestration protocol. Solo path (superpowers executing-plans) — single-file fixes, config, docs, debugging. Team composition is enforced by /team-execute skill, not just advisory. Workers…
We split this CLAUDE.md into the rules it gives an agent
6 rules. About 0 are common, 5 are distinctive, and 1 is very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈5
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 8
Start every non-trivial task with superpowers:brainstorming.
- Distinctive: line 11
When unsure whether a task is exempt, brainstorm.
- Distinctive: line 16
YAGNI: No unnecessary abstractions
- Distinctive: line 18
>=3 options: Before any design decision
- Distinctive: line 56
Prefer a single focused subagent over multiple broad ones
- Distinctive: line 59
Commit context to agents-workbench before ending long sessions.
≈ “Common” is an estimate: at least 1 in 5 of the 25,680 CLAUDE.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
Engineering Standards
Role
Senior engineer mindset. Rigor > speed. Adapt to the project's primary language and stack (edit this line for your domain).
Brainstorm First (required for non-trivial tasks)
Start every non-trivial task with superpowers:brainstorming.
Exempt: typos, comments, reading files, running tests, answering questions.
"Just do it" runs a quick brainstorm. "Skip brainstorm" truly skips.
When unsure whether a task is exempt, brainstorm.
Principles
- Atomicity: >1 concern → break down first
- No placeholders: Complete code only
- YAGNI: No unnecessary abstractions
- Verify before claiming: any response asserting task completion must contain the output of a verification command (test, build, or linter appropriate to the change) in that same response
- >=3 options: Before any design decision
- Security > Correctness > Performance > Style
Workflow
brainstorm → plan (includes execution method) → execute → verify → PR → review → merge
Execution Model
Two paths, chosen during planning (writing-plans must set "Execution method: solo | team"):
Team path (/team-execute) — >=2 source files or design decisions:
- Principal Engineer: architecture, language/platform conventions, security review
- QA Engineer: test quality, mutation testing, integration verification, PR readiness
- Workers: implementation in isolated worktrees (TDD enforced by
tdd-guard.shhook on all Write/Edit, both team and solo paths) - See agents/ for role definitions. See /team-execute for orchestration protocol.
Solo path (superpowers executing-plans) — single-file fixes, config, docs, debugging.
Team composition is enforced by /team-execute skill, not just advisory.
Workers produce draft PRs. QA promotes to ready-for-review after all gates pass. The Principal Engineer can reject and send back with specific fix requests.
TDD Protocol (enforced by hook)
Red → Green → Refactor, in order. Tests are contracts: if a test fails, fix the implementation (unless the test has a genuine bug). Change tests and implementation in separate turns; commit them in separate commits. See /tdd-protocol for full details.
Worktree Workflow (enforced by hook)
Implementation happens in worktrees. agents-workbench is read-only for source code; writes are blocked by enforce-worktree.sh.
See /worktree-guide for commands and flow.
Iteration Budget
Trivial:1 | Simple:2 | Moderate:3 | Complex:4 iterations before escalating to the user.
Subagent Discipline
- Agent teams: parallel teammates (each in own worktree)
- Launch regular (non-team) subagents one at a time
- Prefer a single focused subagent over multiple broad ones
Context Hygiene
Commit context to agents-workbench before ending long sessions.
Domain Knowledge
NOTE: rules/ files are auto-loaded by Claude Code. Do NOT @-import them here.
Doing so would double-load and waste tokens.
See rules/ for: constitution, go-conventions, k8s-conventions, container-conventions,
git-workflow, security, learned-anti-patterns.
In this repository48 other files it gives its agents.
- .claude/templates/AGENTS.mdAGENTS.md
- docs/agents.mdAGENTS.md
- .cursor/rules/core.mdcCursor rule
- .cursor/rules/go.mdcCursor rule
- .cursor/rules/k8s.mdcCursor rule
- .cursor/rules/node.mdcCursor rule
- .cursor/rules/python.mdcCursor rule
- .cursor/rules/rust.mdcCursor rule
- .cursor/rules/tdd.mdcCursor rule
- .cursor/rules/workbench.mdcCursor rule
- llms-full.txtllms.txt
- llms.txtllms.txt
- cmux-sprint-orchestratorSkill
- config-auditSkill
- doneSkill
- eurekaSkill
- goalSkill
- go-reviewSkill
- handoffSkill
- js-reviewSkill
- k8s-debugSkill
- k8s-reviewSkill
- kickoffSkill
- pr-review-ingestSkill
- pr-reviewSkill
- python-reviewSkill
- reflectionSkill
- skill-evalSkill
- tdd-protocolSkill
- team-executeSkill
- team-planSkill
- team-shutdownSkill
- test-reviewSkill
- validate-recommendationSkill
- worktree-guideSkill
- babysitSkill
- canvasSkill
- create-hookSkill
- create-ruleSkill
- create-skillSkill
- create-subagentSkill
- migrate-to-skillsSkill
- sdkSkill
- shellSkill
- split-to-prsSkill
- statuslineSkill
- update-cli-configSkill
- update-cursor-settingsSkill
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.

