agentleFS
kabo-shkabo-sh/kabo-mcp-plugin/AGENTS.md

kabo-mcp-plugin

AGENTS.md2 starsChanged 20 days agoMITView on GitHub

- Everything in this public repository is written in English: commit messages, PR titles and PR descriptions, code comments, docstrings, test names, CHANGELOG.md, release notes, maintainer documentation in docs/, plugin metadata and user-facing content. docs/i18n/ holds the localized README translations and is the only directory where prose in another language is allowed. Outside it, non-English text may appear only as data: localized report strings, patterns that match non-English user input or platform content, and test fixtures that exercise them. scripts/repo-rules.test.mjs…

The file

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

4 rules. About 0 are common, 2 are distinctive, and 2 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.

  • 2

    Very specific

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

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

    Follow docs/skills-spec.md when changing a Skill.

    line 20
  2. Distinctive:

    Default builds and verification read only this repository.

    line 22
  3. Distinctive:

    Add no other hooks. Save complete MCP responses explicitly and verify digests, run ownership and reports.

    line 39
  4. Very specific:

    Use Node 24 to run npm ci --ignore-scripts and npm run verify.

    line 45

≈ “Common” is an estimate: at least 1 in 5 of the 41,581 AGENTS.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
AGENTS.md

Kabo plugin repository

  • Everything in this public repository is written in English: commit messages, PR titles and PR descriptions, code comments, docstrings, test names, CHANGELOG.md, release notes, maintainer documentation in docs/, plugin metadata and user-facing content. docs/i18n/ holds the localized README translations and is the only directory where prose in another language is allowed. Outside it, non-English text may appear only as data: localized report strings, patterns that match non-English user input or platform content, and test fixtures that exercise them. scripts/repo-rules.test.mjs enforces this; a file that needs such data must be listed in its allowlist, and its comments, docstrings and messages stay in English (a comment may quote the literal token it is about) scripts/pr-language-check.mjs (workflow PR language) fails a pull request whose title, description or commit messages contain Chinese, Japanese or Korean prose.
  • The repository root is the plugin root. Maintain one skills/ directory, one kabo entry and one build entry point. Codex, Claude Code and Cursor add only host metadata; do not copy business content or remove features for a particular host.
  • The user selected twenty-two Skills; see skills.lock.json. Eighteen were adapted from creator-skills and are now edited directly in this repository; four more were written here: three for the creator's daily routine (start-with-kabo, pick-daily-viral-videos, review-new-posts) and review-thumbnail, which reviews uploaded images and calls no Kabo data tool. Do not overwrite any of them with a new creator-skills import. The lock file records only the starting point of the adaptation. Follow docs/skills-spec.md when changing a Skill. Enforcement stays off: keep enforced in skills-spec.json empty and do not add entries. Default builds and verification read only this repository. They do not read neighboring repositories or download Skills at install or run time.
  • The Kabo service, not this repository, owns the API, OAuth, database, web app and deployment. Research judgments, scripts and reports run in the AI host. The plugin runs independently of Kabo's web agent and never calls it, directly or through the API, a queue or another proxy. They share no sessions, tasks, model loop, sandbox or report chain; only the Skills share an origin.
  • A multi-round fetch runs in one backend data tool, not as rounds handed to the host model (docs/skills-spec.md KS9, enforced by scripts/fetch-orchestration.test.mjs). Judgment needed between fetch rounds may run as a backend prompt node inside that tool; analysis and the report stay in the host. Scripts never branch on the host to choose a fetch path.
  • The only production MCP connection is /mcp-pub. Each data operation is pinned and must match the provider's artifacts. Do not introduce a registry, a general-purpose executor or remote end-to-end research tasks. The provider keeps the old dynamic server protocol compatible.
  • One lifecycle hook only: hooks/hooks.json runs SessionStart (startup) in Claude Code to offer start-with-kabo once after install, because the host has no install-time event. It only adds a note to the model's context: it never starts a Skill, calls Kabo, or saves results for the model. It is not a package component, so the release ZIP excludes it and the OpenAI package check still rejects hooks/. Add no other hooks. Save complete MCP responses explicitly and verify digests, run ownership and reports; do not read host session records. The static package version and the source Skills' business versions are maintained separately, and packaged content never updates itself at run time.
  • The package must not contain apps, .app.json, a private registry, real local paths, test URLs or credentials. The host manages user authentication. A public signature-verification key may be packaged, but a content digest must not be presented as a digital signature or server attestation.
  • Use Node 24 to run npm ci --ignore-scripts and npm run verify; use Python 3 to check the static scripts. A release ZIP reads only a full commit and must not include working-tree changes. Every Skill resource must be packaged in full.
  • Code goes through PR review; never push directly to main. Do not deploy automatically, upload to a marketplace console or Publish. CI, a buildable ZIP, real-host and production acceptance, and marketplace review are separate facts. State any missing acceptance explicitly and never claim the plugin is ready for listing.
  • Kabo Web serves the Skills of the latest GitHub release (desktop stays on its frozen channel). Publishing a release runs .github/workflows/web-publish.yml, which packs each Skill and rewrites the Web overlay (scripts/web-publish/plugin-web.mjs; it never writes a channel index) when the web-publish environment variable PLUGIN_WEB_AUTO is on; PLUGIN_WEB_HOLD keeps listed Skills on their Registry version in production (the pre test target ignores it; PLUGIN_WEB_TEST_HOLD is its own list). A Skill on no channel and not yet on the overlay is skipped unless it is in PLUGIN_WEB_DEBUT (both targets), in which case it debuts at its SKILL.md metadata.version (required); hold wins over debut, and listing an already-live Skill has no effect. To publish by hand, or to the pre test accounts, run gh workflow run web-publish.yml -f tag=<tag> -f target=index|test (idempotent; a no-op when Web is already on the release). Check with curl -s https://agent.kabo.sh/health | jq '.skill_channels.production.runtime.plugin_overlay'.

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.