claude-kit
CLAUDE.md2 starsChanged 7 days agoMITView on GitHub
Working on the kit itself. Build, test, and release steps live in README.md "Develop"; this file holds only what that doesn't. Every behavior the kit applies everywhere has exactly one owner, listed below. Before adding or changing guidance, find its owner here: change it there, or nowhere. Every other file links to the owner or says nothing about the topic. It never restates, paraphrases, or reminds.
In this repository37 other files it gives its agents.
- templates/CLAUDE.mdCLAUDE.md
- auditSkill
- backup-patternsSkill
- bash-patternsSkill
- depsSkill
- django-patternsSkill
- docker-patternsSkill
- drf-patternsSkill
- engineering-fundamentalsSkill
- fastapi-patternsSkill
- git-patternsSkill
- go-patternsSkill
- graphql-patternsSkill
- investigateSkill
- javascript-patternsSkill
- logging-patternsSkill
- referenceSkill
- metaSkill
- monitoring-patternsSkill
- next-app-router-patternsSkill
- nuxt-patternsSkill
- opentofu-patternsSkill
- prSkill
- python-patternsSkill
- react-patternsSkill
- report-formatSkill
- security-patternsSkill
- slop-patternsSkill
- storybook-patternsSkill
- tailwind-patternsSkill
- test-patternsSkill
- typescript-patternsSkill
- unslopSkill
- vps-provisioningSkill
- vue-patternsSkill
- wcag-auditSkill
- writeSkill
The file
We split this CLAUDE.md into the rules it gives an agent
2 rules. About 0 are common, 2 are distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈2
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 7
Every behavior the kit applies everywhere has exactly one owner, listed below.
- Distinctive: line 10
When two files disagree, the owner wins and the other file is fixed.
≈ “Common” is an estimate: at least 1 in 5 of the 26,980 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
claude-kit
Working on the kit itself. Build, test, and release steps live in README.md "Develop"; this file holds only what that doesn't.
One source of truth
Every behavior the kit applies everywhere has exactly one owner, listed below. Before adding or changing guidance, find its owner here: change it there, or nowhere. Every other file links to the owner or says nothing about the topic. It never restates, paraphrases, or reminds.
- Data and mechanics belong to code:
kit.ymland thekitbinary. Prose never re-describes what a subcommand does. - When two files disagree, the owner wins and the other file is fixed.
- A topic missing from this table gets a row before it gets text anywhere else.
| Topic | Owner |
| Reply length, shape, where output goes, writing files with Edit/Write, code-block tags, external-communication tone, review comment shape | rules/output.md |
| Evidence, confidence labels, never invent, stop after three failed fixes | rules/evidence.md |
| Minimal change, precedent, blast radius, dead code, comments, abstractions, simplification criteria, less code | rules/change.md |
| Which pattern a size, spread, or repetition signal calls for | skills/engineering-fundamentals/reference/design-patterns.md |
| Size, nesting, branch, and coupling thresholds | skills/engineering-fundamentals/reference/metric-thresholds.md |
| Done and verification | rules/verify.md |
| Git and commit policy, side effects, shell mechanics, plans and the step pause, ask vs proceed, AskUserQuestion, review findings, which skills are procedure skills | rules/workflow.md |
| Subagent use and agent defaults | rules/subagents.md |
| CLI choice, read budget, calling bin scripts, scratch and plan location | rules/tooling.md |
| Artifact naming and path | kit scratch-dir |
| Resolving a linked ticket or doc before work starts | skills/investigate step 0 |
| Report shape and severity | skills/report-format |
| Test design | skills/test-patterns |
| Red-green-refactor cycle | skills/engineering-fundamentals/reference/red-green-refactor.md |
Guard lists, skill maps, deny lists and flags, gate env and timeout, on_edit formatting, the CLI tools list, the AI-tell lexicon and its tiers | kit.yml |
| Which checks, fixers, hooks, and package managers a project has | its own configs, read by go/internal/sources; gap-fill (go/internal/gapfill) for what they leave open |
| What runs at each checkpoint: full gate, Stop, per edit, commit message | go/internal/enforce |
| Languages a project uses, and its subprojects | the manifests go/internal/sources reads (Languages, Anchors) |
| Where a tool's binary comes from, and when PATH is allowed | go/internal/resolve |
| How task bodies and CI steps are read | go/internal/classify |
Finding AI tells in text, and the unslop: allow marker | go/internal/slop |
| What an AI tell is and how to rewrite one, in prose and in UI | skills/slop-patterns |
| Which skills and rules the claude.ai chat bundle carries | kit.yml chat_bundle: |
| How the chat bundle is built from them | go/internal/chatbundle |
| Personal register and typography | the user's own ~/.claude/rules/voice.md, never the kit |
Skills and agents carry procedure and stack knowledge only. When one needs a rule above, it cites rules/<file>.md instead of restating it.
Run the checkout's kit
In this repo, call bin/kit, never bare kit: on PATH, kit is the installed release, which plans and checks with the old code. With KIT_DEV=1, bin/kit builds from go/. This overrides rules/tooling.md section 2 here only.
Rules are live
The SessionStart and SubagentStart hooks inject rules/*.md from the plugin root they run from. A session running the hooks from this checkout picks up a rule edit at its next start or compaction, so finish a rules change in one sitting.
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.