anything
AGENTS.md1 starsChanged 9 days agoApache-2.0View on GitHub
Anything is a lightweight collection of opinionated, independently useful agent skills for software, research, writing, planning, and other work. Keep useful defaults available without setup. Setup and cleanup are explicitly user-invoked. Setup records durable project preferences in AGENTS.md; it must preserve existing instructions and avoid baking in the current harness or environment. This is an instruction-only repository. There is no required build, runtime dependency installation, or application test suite. For relevant changes, check JSON and YAML metadata, names and versions,…
Where files are headingNew AGENTS.md files have more rules about keeping secrets out of the code, and fewer about code style.
Of the AGENTS.md files we read, the 32,671 first seen in the last six months, against the older ones: the kind of rule that gained most share, and the kind that lost most. About files like this in general, not this one, and an estimate, since a kind is a pattern in the words.
In this repository12 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- approval-gateSkill
- cleanupSkill
- fresh-reviewSkill
- git-styleSkill
- interviewSkill
- no-tombstonesSkill
- orchestrateSkill
- repo-styleSkill
- setupSkill
- task-planSkill
- using-anythingSkill
The file
We split this AGENTS.md into the rules it gives an agent
9 rules. About 0 are common, 4 are distinctive, and 5 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈4
Distinctive · estimate
This author's own choices. Worth a look.
5
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 7
Keep useful defaults available without setup.
- Distinctive: line 11
Keep behavioral instructions based on capabilities and permissions, not named agent tools.
- Distinctive: line 17
Keep changes focused. Maintain existing names and organization unless changing them solves a concrete problem.
- Very specific: line 23
Keep setup and cleanup's manual-invocation frontmatter and their agents/openai.yaml policies intact.
- Very specific: line 24
Keep names and versions consistent in plugin.json, .claude-plugin/plugin.json, and package.json.
- Very specific: line 26
Use Apache-2.0 licensing metadata.
- Very specific: line 28
Use README.md for the entry point and docs/ for substantial installation, defaults, and compatibility detail.
- Distinctive: line 37
Run git diff --check before committing.
- Very specific: line 50
Use Conventional Commits for normal development, with focused commits and configured author identity.
≈ “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.
Files
Working on anything
Purpose
Anything is a lightweight collection of opinionated, independently useful agent skills for software, research, writing, planning, and other work.
Keep useful defaults available without setup. Setup and cleanup are explicitly user-invoked. Setup records durable project preferences in AGENTS.md; it must preserve existing instructions and avoid baking in the current harness or environment.
Authoring
- Keep behavioral instructions based on capabilities and permissions, not named agent tools. Put host-specific installation and invocation details in metadata or documentation.
- Delegation is optional. Support direct execution, and describe subprocess workers only when tooling and lifecycle controls are available and their use is permitted. Never use a fallback to bypass a restriction.
- Make every skill useful on its own. Keep required templates, references, licenses, and notices inside that skill's directory; sibling skills and root repository resources are optional.
- Match planning, questions, verification, and documentation to the task. Avoid mandatory agent counts, Git operations, approval rituals, or document sets for all work.
- Distinguish observations from assumptions and independent review from self-review. Report actual checks and limitations.
- Respect existing project conventions and authorization within their scope. Ask about consequential unknowns or new scope, not decisions already established.
- Keep changes focused. Maintain existing names and organization unless changing them solves a concrete problem.
Files and metadata
- Each skill lives in
skills/<name>/SKILL.md, with a matching lowercase kebab-case frontmatter name and a clear description of when it applies. - Supporting paths are relative to the skill directory.
- Keep setup and cleanup's manual-invocation frontmatter and their
agents/openai.yamlpolicies intact. - Keep names and versions consistent in
plugin.json,.claude-plugin/plugin.json, andpackage.json. Keep both marketplace catalogs aligned with the plugin identity. - Bump the shared version whenever skill instructions, bundled supporting files, or plugin behavior or packaging change. Related edits in one release share one bump; repository-only documentation changes do not require one. Use a patch for corrections, a minor for new skills or compatible capabilities, and a major for breaking changes. See release versioning for the update and cache behavior.
- Use
Apache-2.0licensing metadata. Keep each skill'sLICENSE.txtandNOTICEidentical to the rootLICENSEandNOTICEso individual installations retain the terms and attribution. assets/holds the wordmark and plugin icon, drawn as outlines in ink#111111and paper#eeeeee, so no font is required. The README and manifests reference these files; skills must not depend on them. See logos and icons for each host's field.- Use
README.mdfor the entry point anddocs/for substantial installation, defaults, and compatibility detail. Update the relevant documentation when behavior changes. Do not add a repository changelog. CLAUDE.mdimports this file; maintain project instructions here rather than duplicating them.
Verification
This is an instruction-only repository. There is no required build, runtime dependency installation, or application test suite.
For relevant changes, check JSON and YAML metadata, names and versions, local Markdown links, self-contained skill references, and the affected cases in docs/compatibility.md. Check that skill instructions and examples agree.
Run git diff --check before committing. When the CLI is available, validate the native Claude manifests:
claude plugin validate .claude-plugin/plugin.json --json
claude plugin validate .claude-plugin/marketplace.json --strictExpect the plugin validator's warning that root CLAUDE.md is not loaded as installed plugin context: it is deliberately for contributors. Check that the report has no errors, and investigate any other manifest or content warnings. Strict mode also rejects that intentional warning.
These validate packaging, not end-to-end agent behavior. Do not claim runtime testing in a host that was not exercised. Keep validation from changing installed plugins as a side effect.
Git
Use Conventional Commits for normal development, with focused commits and configured author identity. The initial public snapshot intentionally uses Initial commit.
Preserve unrelated work. Commit, publish, or rewrite history within the user's authorized scope, respecting applicable restrictions. Do not add a repeated approval step to an action already authorized.
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.

