agentleFS
azamkhanukazamkhanuk/context-central/.claude/CLAUDE.md

context-central

CLAUDE.md2 starsChanged 30 days agoMITView on GitHub

A Claude Code plugin: a CLI in lib/, two hooks, seven skills and three agents. Terms are defined in CONTEXT.md; use them as written. This repository is public-safe. Nothing in it, and no commit message, may name a client, an employer, a private person, a real ticket or a private path. Public products the plugin works with, such as GitHub and Jira, may be named. The one person named is the author: in LICENSE, in the two manifests and the…

  • Reads credentials
  • Installs packages
  • Commits and pushes

The file

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

4 rules. About 0 are common, 0 are distinctive, and 4 are 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.

  • 4

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Before any push: node scripts/scan-names.mts <names-file>.

    line 10
  2. Very specific:

    Every tracked text file is LF, pinned in .gitattributes.

    line 18
  3. Very specific:

    No lockfile is committed. With a package.json and a lockfile at the plugin root, Claude Code runs an npm install in every user's plugin …

    line 34
  4. Very specific:

    Test through the CLI as a process: run() and hook() in test/helpers.mts, against a tree made with acme() or makeTree().

    line 45

≈ “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

Working on context-central

A Claude Code plugin: a CLI in lib/, two hooks, seven skills and three agents. Terms are defined in CONTEXT.md; use them as written.

Checks

  • One test file: node --test test/<name>.test.mts. All of them: npm test.
  • Types: npm run typecheck. It must stay clean.
  • Manifests: claude plugin validate .. It must pass.
  • Before any push: node scripts/scan-names.mts <names-file>. It must exit 0.

Public-safety rule

This repository is public-safe. Nothing in it, and no commit message, may name a client, an employer, a private person, a real ticket or a private path. Public products the plugin works with, such as GitHub and Jira, may be named. The one person named is the author: in LICENSE, in the two manifests and the test that pins them, and as the GitHub account in the install commands and in the marketplace's name and address, which lib/templates.mts holds and the tests pin. Fixtures use the invented estate "acme" and keys like PROJ-12. The list of names to keep out lives outside the repository and is never committed.

Conventions

  • Every tracked text file is LF, pinned in .gitattributes.
  • TypeScript in .mts files, checked in strict mode and run by Node as it is: nothing is compiled, and tsc only checks. So only syntax Node can strip: no enum, no namespace, no parameter properties, import type for a type, and imports that end in .mts.
  • No any and no @ts- comments. A cast only where data comes in from outside: parsed JSON, flags, a caught error, the environment. A type lives beside the code that makes the thing.
  • Node 22.18 or later, Node built-ins only. bin/context-central is the one JavaScript file: it has to run on a Node too old for the rest, to say so. Flags are parsed with parseArgs from node:util.
  • A command is lib/commands/<name>.mts exporting summary and run(args, io). The router finds it by file name; nothing registers it.
  • io is { cwd, env, nodeVersion, out, err, readStdin }. Commands never touch process directly.
  • Expected failures throw PluginError (exit 1) or UsageError (exit 2) from lib/errors.mts. Exit 0 means fine.
  • --json on commands whose output a skill or script parses.
  • The clock is localDate(io.env); tests fix it with CONTEXT_CENTRAL_NOW. The map lock alone reads the system's clock, since a lock folder's age is the file system's own time, which CONTEXT_CENTRAL_NOW cannot move.
  • Config paths are POSIX, relative to the map directory (nodes) or the estate root (repos, hub, a repo's standards).
  • The plugin holds no rule for any language, framework or build tool, in lib/, a skill or an agent. How a repo's code is written and checked comes from its own files, its standards note and its recorded checks.
  • Hooks always exit 0 and print nothing when they have nothing to say. What they add is facts, never instructions.
  • Skills set neither model nor effort; estate facts come from context-central config --get, never from skill text.
  • A skill has a file of the map read and written at the path context-central draft prints, and put into the map by context-central take, never at a path it works out for itself: in a git work tree a host may keep a session's tools to the work tree.
  • plugin.json carries the version and the marketplace entry carries none. People stay on a release until the version changes, so a change reaches them only through a release.
  • The plugin installs from its author's marketplace, a repository of its own that lists it. lib/templates.mts holds that marketplace's name and address, and a test holds the README and the install doc to them. The repository's own marketplace, named after the plugin, is for installing from a clone and for the copies installed from it before 0.14.0.
  • No lockfile is committed. With a package.json and a lockfile at the plugin root, Claude Code runs an npm install in every user's plugin cache, and the plugin has no runtime dependencies.

Releases

  • A release is one pull request that raises version in .claude-plugin/plugin.json and adds that version's section, dated, at the top of CHANGELOG.md. A test holds the two together.
  • Once it is merged and the checks pass on main, .github/workflows/release.yml tags that commit context-central--v<version> and publishes the GitHub release, with that section of the changelog as its notes. scripts/release-notes.mts prints the tag, the title and the notes. A merge that does not raise the version publishes nothing.
  • By hand, the same is claude plugin tag --push on main, then gh release create <tag> --verify-tag.

Tests

  • Work test-first, one behaviour at a time.
  • Test through the CLI as a process: run() and hook() in test/helpers.mts, against a tree made with acme() or makeTree(). Never import a module under test.
  • Expected values are literals or come from the fixture, never from re-running the code's own logic.
  • A skill or an agent is tested where it meets the CLI: its frontmatter, and each command, flag, setting, file or printed line it names. Its wording is not pinned by a test: what a session does with it is shown by a run.
  • The checks run on Linux, macOS and Windows. Build an expected path with join. A test that cannot run on a system is skipped there with its reason, through notOnWindows or onlyOnWindows in test/helpers.mts.

Writing

  • Code needs no comments. One line only where the code cannot hold the reason. None in tests.
  • Text a person or the model reads is plain British English, with no em dashes and no emojis.
  • The README is the short version: what the plugin does, the install and the five steps. The detail of each part is in its own file under docs/, which the README links, and new detail goes there. A test holds the links.

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.