agentleFS
theishandubeytheishandubey/dotclaude/claude/CLAUDE.md

dotclaude / claude

CLAUDE.md2 starsChanged 27 days agoMITView on GitHub

These apply to every session and agent, in every project. Routing and dispatch rules for the main session live in the orchestrator playbook (orchestrator.md), which only the main session receives; keep them out of this file. - Never use an em dash; use a plain dash "-". - Do not write comments or docstrings in code, in any language. Make the code carry the reason instead, in this order: a name, a small helper or a type; then the commit…

The file

Show its rules and text

We split this CLAUDE.md into the rules it gives an agent

21 rules. About 0 are common, 17 are distinctive, and 4 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈17

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 4

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  • could be automatic: a setting could enforce it
  1. Distinctive:

    Never use an em dash; use a plain dash "-".

    line 8
  2. Distinctive:

    Do not write comments or docstrings in code, in any language.

    line 9
  3. Distinctive:

    Only when neither can show a constraint the code depends on (a deadlock, a platform or library quirk, an external API's behavior) write one …

    line 11
  4. Distinctive:

    Never write comments that say what the code does, section headers, TODOs or commented-out code.

    line 12
  5. Distinctive:

    When you change code, delete the comments this rule does not allow in the lines you touch, and move any reason they carried into the places …

    line 13
  6. Distinctive:

    Never edit CHANGELOG.md files or files marked as auto-generated.

    line 15
  7. Very specific:

    Only implementer, parallel-implementer and fixer write and run tests, following their preloaded testing skill: unit tests for every change, …

    line 21
  8. Distinctive:

    Treat lint errors, type errors and build failures the same way, whether or not your change caused them: fix them when your role and scope …

    line 25
  9. Very specific:

    Call only the tools in your own tool list.

    line 26
  10. Very specific:

    Put scratch files in the session scratchpad when one is given, otherwise in a mktemp -d directory, never inside a repository, and leave …

    line 27
  11. Distinctive:

    Never work around a block.

    line 29
  12. Very specific:

    Never poll with sleep followed by a read.

    line 30
  13. Distinctive:

    Run a long command in the foreground with a timeout, or in the background and wait for its completion notice.

    line 31
  14. Distinctive:

    Default to subtraction: remove dead code, redundant validators and stubs before building on the simpler base.

    line 35
  15. Distinctive:

    Keep each decision in one place and pass its result along, instead of repeating the same choice in several places.

    line 37
  16. Distinctive:

    When you replace an internal API, migrate every caller and delete the old one in the same change, unless external users depend on it.

    line 38
  17. Distinctive:

    When the same correction or instruction comes up a second time, encode it as structure (a lint, type, check or script) instead of more text.

    line 41
  18. Distinctive:

    Only when it needs judgment, make the instruction more prominent and add an example of the failure.

    line 42
  19. Distinctive:

    Commit messages, branch names and PR titles describe the change only: no plan numbers, milestone labels, task numbers or step references.

    line 47
  20. Distinctive:

    When a change makes a decision the code cannot show, explain it in the commit message body.

    line 49
  21. Distinctive:

    Never add an agent or Claude co-author trailer.

    line 50

≈ “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.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
CLAUDE.md

Shared Rules

These apply to every session and agent, in every project. Routing and dispatch rules for the main session live in the orchestrator playbook (orchestrator.md), which only the main session receives; keep them out of this file.

Writing

  • Never use an em dash; use a plain dash "-".
  • Do not write comments or docstrings in code, in any language. Make the code carry the reason instead, in this order: a name, a small helper or a type; then the commit message body. Only when neither can show a constraint the code depends on (a deadlock, a platform or library quirk, an external API's behavior) write one line saying why. Never write comments that say what the code does, section headers, TODOs or commented-out code. When you change code, delete the comments this rule does not allow in the lines you touch, and move any reason they carried into the places above rather than losing it. Machine-read directives are allowed when the tooling requires them: shebangs, encoding lines, license headers the project mandates, and linter, type-checker or compiler pragmas.
  • Never edit CHANGELOG.md files or files marked as auto-generated.
  • In long Markdown files, put each sentence on its own line and keep normal Markdown structure.

Engineering

  • Weigh quality, simplicity, robustness, scalability and long-term maintainability above development cost.
  • Only implementer, parallel-implementer and fixer write and run tests, following their preloaded testing skill: unit tests for every change, end-to-end tests only when the user asks for them. Everyone else does not write, run or review tests, and verifies a change with the project's build, type-checker or linter, or by running the changed command once. Code and security review belong to the user.
  • Fix a bug only after reproducing it with the cheapest command that shows the symptom.
  • Treat lint errors, type errors and build failures the same way, whether or not your change caused them: fix them when your role and scope allow, otherwise report them with the exact output so they get routed.
  • Call only the tools in your own tool list; the shell tool is Bash.
  • Put scratch files in the session scratchpad when one is given, otherwise in a mktemp -d directory, never inside a repository, and leave them for the OS to clean.
  • A guard hook blocks catastrophic commands (wiping the filesystem, home or project, dropping databases, destroying infrastructure, force-pushing main). Never work around a block; tell the user what you wanted to run.
  • Never poll with sleep followed by a read. Run a long command in the foreground with a timeout, or in the background and wait for its completion notice.

Principles

  • Default to subtraction: remove dead code, redundant validators and stubs before building on the simpler base. Make the smallest change that solves the problem, and add no abstraction, layer, guard or option the current need does not demand.
  • Keep each decision in one place and pass its result along, instead of repeating the same choice in several places.
  • When you replace an internal API, migrate every caller and delete the old one in the same change, unless external users depend on it.
  • In a planned rewrite or migration, converge on the target design instead of keeping every intermediate step working with throwaway compatibility code. Declare where temporary breakage is acceptable and verify everything at the end.
  • When the same correction or instruction comes up a second time, encode it as structure (a lint, type, check or script) instead of more text. Only when it needs judgment, make the instruction more prominent and add an example of the failure.

Commits and branches

  • Make atomic commits: one logically complete change each, each building and passing on its own.
  • Commit messages, branch names and PR titles describe the change only: no plan numbers, milestone labels, task numbers or step references. Name a branch for what it changes, in the repo's branch convention or short kebab-case: add-top-limit or fix-login-redirect, never plan-003, milestone-2 or step-4.
  • When a change makes a decision the code cannot show, explain it in the commit message body.
  • Never add an agent or Claude co-author trailer.

Download now (ZIP)

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
What did you use it for? e.g. Reviewing pull requests in a Rails app

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.