nuvek-agent-config
AGENTS.md0 starsChanged 17 days agoMITView on GitHub
Contributor rules for this repository. It is the single authored source of instructions for Claude Code, Antigravity and Codex. A tool's own configuration is a symlink into this repository or a file generated from it, so edit here and never edit a projection. - .claude-plugin/plugin.json and skills/<name>/SKILL.md: the plugin and its skills. - rules/core.md: shared mandates, in tool-neutral wording. rules/claude.md, agy.md and codex.md are per-tool overlays. - sync/: neutral sources (agents/, commands/, mcp.toml, permissions.toml, hooks.toml, map.toml, optional models.toml) that agent-sync…
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 31,700 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 repository9 other files it gives its agents.
- rules/claude.mdCLAUDE.md
- delegationSkill
- git-workflow-expertSkill
- package-releaseSkill
- polyglot-architectSkill
- rust-craftsmanSkill
- security-governanceSkill
- shared-library-reviewSkill
- skill-creatorSkill
The file
We split this AGENTS.md into the rules it gives an agent
2 rules. About 0 are common, 0 are distinctive, and 2 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.
2
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 27
One skill, one job: If a skill covers two unrelated jobs, split it.
- Very specific: line 35
No private facts: No hostnames, domains, addresses, usernames or home paths.
≈ “Common” is an estimate: at least 1 in 5 of the 40,246 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
AGENTS.md
Contributor rules for this repository. It is the single authored source of instructions for Claude Code, Antigravity and Codex. A tool's own configuration is a symlink into this repository or a file generated from it, so edit here and never edit a projection.
Layout
.claude-plugin/plugin.jsonandskills/<name>/SKILL.md: the plugin and its skills.rules/core.md: shared mandates, in tool-neutral wording.rules/claude.md,agy.mdandcodex.mdare per-tool overlays.sync/: neutral sources (agents/,commands/,mcp.toml,permissions.toml,hooks.toml,map.toml, optionalmodels.toml) thatagent-sync applyrenders per tool.*.examplefiles are templates and are not read byagent-sync.
Never add agents/, commands/ or hooks/ at the repository root. The root is also the Claude plugin
root, and Claude loads those directories there as plugin components.
Rules for editing skills
- Neutral wording. Name model tiers (
fast,standard,deep), "the interactive question tool", MCP tools asserver→tool, and CLIs by their bare name. A passage that only one tool can follow goes in its overlay or inside an<!-- only:claude -->or<!-- only:agy -->block. - Present-state facts. A skill states how things are now: the rule, the failure mode and the fix. No dates, no changelog prose, no "used to", "was" or "deprecated". Something that no longer exists gets no mention.
- One skill, one job. If a skill covers two unrelated jobs, split it. Resolve overlap between skills in
their descriptions and in the routing section of
rules/core.md. - How, not what. A skill teaches procedures and failure modes. It never stores facts that change on their own (inventories, addresses, versions, file lists).
- Strict YAML frontmatter. Quote any
descriptionthat contains:. Claude tolerates sloppy frontmatter; Antigravity drops the whole skill.nameequals the folder name. - Optional integrations stay optional. A skill may name
cross-agent,hive-boardor another tool, but it must still work without it and say what to do instead. - No private facts. No hostnames, domains, addresses, usernames or home paths. Use a placeholder or a
<!-- customise: ... -->marker. CI greps for these and fails on a hit. - No agent artefacts. Plans, reviews, QA notes and investigation notes do not belong in this repository, and commit messages and comments do not cite them.
- Canadian English in prose. Code identifiers keep their own spelling.
Before you commit
agent-sync check --dir "$PWD"against a scratchHOMEmust parse without an error (drift on a clean home is expected).scripts/check-public.shfinds no private deployment details.- The pre-commit hook (
git config core.hooksPath .githooks) runs gitleaks. Never use--no-verify. - Conventional Commits, commit straight to
masterin a personal fork, and sign off (-s) for a pull request.
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.