gars-claude-skills / notes
AGENTS.md1 starsChanged 59 days agoMITView on GitHub
Location: agents/, one flat file per agent (agents/<name>.md), mirroring how Claude Code loads them from ~/.claude/agents/ or a project's .claude/agents/. Different shape from skills/: agent frontmatter carries name, description, tools, and model, and there is no per-agent subfolder or references/. Five agents forming a permission-gated pipeline for shipping code changes: - code-writer - implements the change: reads, edits, writes files. Applies the code-philosophy skill before any non-trivial edit. - code-reviewer - read-only review of a diff or existing files against…
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 31,542 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 repository10 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- claude-project-structureSkill
- code-philosophySkill
- writing-documentationSkill
- fact-checkingSkill
- product-researchSkill
- professional-communicationSkill
- skill-creationSkill
- skill-refactoringSkill
- tooling-researchSkill
The file
We split this agents.md into the rules it gives an agent
3 rules. About 0 are common, 2 are distinctive, and 1 is 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.
1
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 14
code-reviewer: - read-only review of a diff or existing files against the same code-philosophy rubric.
- Distinctive: line 16
code-tester: - read-only: runs the project's actual test command, diagnoses failures, reasons about coverage gaps.
- Distinctive: line 18
code-auditor: - read-only security review of existing code or of a specific dependency someone wants to install.
≈ “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
Coding agents
Location: agents/, one flat file per agent (agents/<name>.md), mirroring how
Claude Code loads them from ~/.claude/agents/ or a project's .claude/agents/.
Different shape from skills/: agent frontmatter carries name, description,
tools, and model, and there is no per-agent subfolder or references/.
What's here
Five agents forming a permission-gated pipeline for shipping code changes:
- code-writer - implements the change: reads, edits, writes files. Applies
the
code-philosophyskill before any non-trivial edit. - code-reviewer - read-only review of a diff or existing files against the
same
code-philosophyrubric. No Edit/Write access. - code-tester - read-only: runs the project's actual test command, diagnoses failures, reasons about coverage gaps. No Edit/Write access.
- code-auditor - read-only security review of existing code or of a specific dependency someone wants to install. Renders an explicit APPROVE/REFUSE verdict. Has WebFetch/WebSearch to check CVEs and advisories.
- dependency-installer - the only agent allowed to run install commands.
The pipeline
code-writer, code-reviewer, and code-tester never install anything. If any
of them decides a new dependency is needed, they stop and report the package
name, ecosystem, version constraint, and why, rather than reaching for
npm install or equivalent. That report routes to code-auditor, which renders
APPROVE or REFUSE: a recommendation, not a hard block, the user has final say.
dependency-installer acts only once a dependency has both an auditor verdict
and explicit user sign-off, and treats those as two independent gates. An
APPROVE is not itself approval to proceed.
Keeping them current
Same discipline as skills/. See syncing-from-global.md.
Origin
Ported 2026-08-17 from the global agent set with a Minimalism-only pass, rather
than the full skill-creation structure, since these are agent definitions and
do not carry a Principles or Example section in Claude Code's agent format.
Source: logs/08-17-2026.md.
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.