agentleFS
gastonlarap-a11ygastonlarap-a11y/claude-code-setup/global/CLAUDE.md

claude-code-setup / global

CLAUDE.md1 starsChanged 3 months agoMITView on GitHub

The file

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

11 rules. About 0 are common, 8 are distinctive, and 3 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈8

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 3

    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:

    Always answer in Spanish (explanations, summaries, questions).

    line 7
  2. Very specific:

    Every feature ships with its tests.

    line 13
  3. Distinctive:

    Prefer running single tests (the repo's single-test command) over full suites while iterating.

    line 14
  4. Distinctive:

    Never assume: if a request is ambiguous, incomplete, or contradictory, ask targeted questions before acting.

    line 19
  5. Distinctive:

    Counter-propose: if what I asked is incorrect, suboptimal, or a more efficient route exists, say so with the why BEFORE acting, and let me …

    line 20
  6. Distinctive:

    Prefer editing existing files over creating new ones.

    line 23
  7. Distinctive:

    Tools have jobs: read with Read, search with Grep/Glob, change with Edit/Write.

    line 24
  8. Very specific:

    Nothing is enabled globally: no plugin or LSP is on by default.

    line 25
  9. Distinctive:

    Scope contract: before the first edit of a task touching 3+ files, state in ≤5 lines which files you will touch (grouped by area), what is …

    line 26
  10. Very specific:

    After a merge is confirmed, or a local branch shows [gone] upstream → post-merge-cleanup skill (verifies the merge, then prunes and …

    line 33
  11. Distinctive:

    When compacting, preserve: the list of modified files, the exact build/test commands used and their latest results, key architectural …

    line 37

≈ “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

Global instructions

About me

Senior polyglot developer. Main stacks: NestJS (TypeScript/Fastify), Go, Android (Kotlin), React/Next.js, Flutter/Dart. I also work with SQL/NoSQL databases, Docker, Kubernetes, testing at every level, and CI/CD with GitHub Actions. I maintain per-stack template repos under ~/Documents/Git/.

Language

  • Always answer in Spanish (explanations, summaries, questions).
  • Internal reasoning and all research: English. Code, identifiers, code comments, commit messages, and config files: English.

Engineering standards (every project)

  • Code must communicate its intent to the compiler as much as to other developers: encode invariants in types, self-documenting names, small focused functions, clear module boundaries. The most frequent reader is me months later — someone else must understand this code without me.
  • Apply design patterns only when they solve a real problem, and say which pattern and why. No over-engineering, no speculative abstractions.
  • Every feature ships with its tests. Run the repo's lint/typecheck/test commands before declaring work done, and report their real results.
  • Prefer running single tests (the repo's single-test command) over full suites while iterating; run the full suite before declaring done.
  • Respect each repo's existing architecture (its CLAUDE.md / ARCHITECTURE.md); never introduce a competing style.
  • Handle errors explicitly at boundaries; never swallow exceptions or ignore returned errors.

Workflow rules

  • Never assume: if a request is ambiguous, incomplete, or contradictory, ask targeted questions before acting.
  • Counter-propose: if what I asked is incorrect, suboptimal, or a more efficient route exists, say so with the why BEFORE acting, and let me choose between my version and yours; once I pick, execute without re-litigating.
  • Research proactively without being asked: for any third-party API surface, version, or "how do I X" not fully covered by a loaded skill, use the research skill — official sources in English, latest stable, only the relevant section. Never a version or signature from memory.
  • VCS via CLI: use gh for GitHub remotes and az repos / az devops for Azure DevOps remotes (detect from git remote -v). If the required CLI is missing, install it or give me the exact install steps and pause until I confirm.
  • Prefer editing existing files over creating new ones. No README/docs files unless asked — except when creating or setting up a project, which always includes its README (setup-project protocol).
  • Tools have jobs: read with Read, search with Grep/Glob, change with Edit/Write. Bash runs things — never use it to read, write or patch repo files: a shell heredoc re-serializes a whole file where an Edit would have sent a diff.
  • Nothing is enabled globally: no plugin or LSP is on by default. Each repo turns on what its own stack needs in its .claude/settings.json — setup-project maps stack → plugins. A plugin loaded where it does not apply is pure context cost.
  • Scope contract: before the first edit of a task touching 3+ files, state in ≤5 lines which files you will touch (grouped by area), what is out of scope, and your confidence line; wait for the go-ahead. Skip for single-file or already-scoped fixes.
  • Confidence: close every plan, "work is done" report, research answer and config proposal with Confianza: NN% — <what stayed unverified>. Never a bare number: name an unverified assumption, or claim none only when everything really ran. Below 70%, say human review is needed before merging.

Git conventions

  • Authorship rule (absolute): commits, PRs, MRs, issues and their descriptions carry ONLY my identity — no Co-Authored-By trailers, no "Generated with" footers, no AI attribution of any kind, anywhere.
  • Publish only on explicit order: never commit, push, or open PRs/MRs on your own initiative — on any platform (GitHub via gh, Azure DevOps via az repos, or other). Only when I explicitly ask ("sube los cambios" or equivalent); then run the full flow (commit → push → PR/MR) with the CLI detected from git remote -v.
  • Conventional Commits style: feat|fix|refactor|test|chore(scope): message.
  • After a merge is confirmed, or a local branch shows [gone] upstream → post-merge-cleanup skill (verifies the merge, then prunes and branches off a fresh main).

Compact instructions

When compacting, preserve: the list of modified files, the exact build/test commands used and their latest results, key architectural decisions made this session, and any unresolved errors or pending steps.

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.