ai-driven-development
AGENTS.md169 starsChanged 56 days agoMITView on GitHub
This repo is a single-plugin marketplace and a cross-agent skill collection in one. The same skills/ directory at the repo root is read by both npx skills add (the cross-agent CLI) and Claude Code's plugin install (/plugin marketplace add + /plugin install …@ai-driven-development). The plugin sees the repo root as itself via source: "./" in marketplace.json — no nesting, no symlinks, no duplication. When the user asks to commit or push without naming a target branch, commit and push directly…
- Commits and pushes
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 31,843 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 repository23 other files it gives its agents.
- CLAUDE.mdCLAUDE.md
- agentic-readinessSkill
- bug-fix-protocolSkill
- clipboardSkill
- code-that-fits-in-your-headSkill
- fetch-url-as-markdownSkill
- fpf-problem-solvingSkill
- hooks-managementSkill
- installing-cli-toolsSkill
- investigating-repository-historySkill
- maintaining-macos-healthSkill
- maintaining-windows-healthSkill
- mcp-managementSkill
- plan-modeSkill
- plugins-managementSkill
- prompt-engineeringSkill
- refactoring-csharpSkill
- repo-activity-summarySkill
- semantic-scholar-deepSkill
- settings-managementSkill
- skills-managementSkill
- ubiquitous-languageSkill
- windows-qa-engineerSkill
The file
Show its rules and text
We split this AGENTS.md into the rules it gives an agent
14 rules. About 0 are common, 4 are distinctive, and 10 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.
10
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- could be automatic: a setting could enforce it
- Distinctive: line 7
When the user asks to commit or push without naming a target branch, commit and push directly to main.
- Distinctive: line 46
Update the relevant section table in README.md
- Very specific: line 47
Update the skill's own SKILL.md (agent-facing) and README.md (human-facing, optional)
- Very specific: line 48
Update .claude-plugin/plugin.json and .claude-plugin/marketplace.json versions
- Very specific: line 52
Do not create or commit per-skill snapshots or edit-history artifacts, including .skill_snapshots/, .snapshots/, and .skill_edit_log.jsonl.
- Very specific: line 67
When skill/hook/plugin files change:
- Very specific: line 71
Create and push tag: git tag -a vX.Y.Z -m "Version X.Y.Z" && git push origin vX.Y.Z
- Very specific: line 72
Create GitHub release: gh release create vX.Y.Z --title "vX.Y.Z — Title" --notes "Release notes"
- Very specific: line 95
Create skills/<skill-name>/SKILL.md with YAML frontmatter (name, description).
- Very specific: line 96
Add optional README.md, references/, scripts/, assets/ as needed.
- Distinctive: line 97
Update the relevant section table in the root README.md.
- Very specific: line 103
Create hooks/<hook-name>/ with the hook implementation, install.sh, and README.md.
- Very specific: line 104
If the hook ships pre-built binaries, add a release flow (see hooks/balanced-safety-hooks/ as the canonical example: cross-platform Go …
- Distinctive: line 105
Update the Hooks section in the root README.md.
≈ “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
AI-Driven Development — Development Guide
This repo is a single-plugin marketplace and a cross-agent skill collection in one. The same skills/ directory at the repo root is read by both npx skills add (the cross-agent CLI) and Claude Code's plugin install (/plugin marketplace add + /plugin install …@ai-driven-development). The plugin sees the repo root as itself via source: "./" in marketplace.json — no nesting, no symlinks, no duplication.
Git publishing defaults
When the user asks to commit or push without naming a target branch, commit and
push directly to main. If the current branch is main, stay on it; do not
create a feature branch by default. Use another branch only when the user
explicitly requests one.
When to create tags and releases
Only create git tags and GitHub releases when skill, hook, or plugin files change.
Requires a new tag/release
skills/*/— any skill file changes (SKILL.md, scripts, references, assets)hooks/*/— any hook source changes (Go code, install scripts, hook configs).claude-plugin/plugin.json— plugin manifest changes.claude-plugin/marketplace.json— marketplace catalogue changesLICENSE— licence changes
Examples:
- Adding a new skill or hook
- Updating skill instructions in SKILL.md
- Modifying skill scripts
- Changing plugin metadata or version
Does NOT require a new tag/release
Documentation-only changes don't need tagging:
README.md— documentation and use-case updatesAGENTS.mdand theCLAUDE.mdimport shim — repository instructions- Commit messages
- GitHub-specific files (
.github/*)
Rationale: users install specific commits via SHA. Documentation updates don't affect functionality, so they don't need version bumps. This keeps the release history clean.
README maintenance
The repo ships English-only. When you add or remove a skill, or change a skill description:
- Update the relevant section table in
README.md - Update the skill's own
SKILL.md(agent-facing) andREADME.md(human-facing, optional) - Update
.claude-plugin/plugin.jsonand.claude-plugin/marketplace.jsonversions
Generated skill artifacts
Do not create or commit per-skill snapshots or edit-history artifacts, including
.skill_snapshots/, .snapshots/, and .skill_edit_log.jsonl. Git history is
the sole record of skill changes. These generated artifacts are ignored by the
root .gitignore and must not be force-added.
Versioning
Follow semver:
- MAJOR (x.0.0) — breaking changes to skill interfaces, removals, layout overhauls
- MINOR (X.x.0) — new skills, new hooks, backward-compatible features
- PATCH (X.X.x) — bug fixes, prompt tweaks, doc fixes that ship in skills
Release process
When skill/hook/plugin files change:
- Bump version in
.claude-plugin/plugin.jsonAND.claude-plugin/marketplace.json(keep them in sync). - Commit changes.
- Create and push tag:
git tag -a vX.Y.Z -m "Version X.Y.Z" && git push origin vX.Y.Z - Create GitHub release:
gh release create vX.Y.Z --title "vX.Y.Z — Title" --notes "Release notes"
The bash-guard hook ships its own pre-built binary release with a separate tag (bash-guard-vX.Y.Z) — see hooks/balanced-safety-hooks/ for that release flow.
Repository structure
ai-driven-development/
├── .claude-plugin/
│ ├── marketplace.json ← single-plugin marketplace, source: "./"
│ └── plugin.json
├── skills/ ← canonical Agent Skills layout (npx skills add)
│ └── <skill-name>/SKILL.md (+ optional README.md, references/, scripts/, assets/)
├── hooks/
│ └── <hook-name>/ ← Claude Code hooks (Go binary + install scripts)
├── README.md
├── AGENTS.md ← canonical repository instructions
├── CLAUDE.md ← imports AGENTS.md for Claude Code
└── LICENSEAdding a new skill
- Create
skills/<skill-name>/SKILL.mdwith YAML frontmatter (name,description). - Add optional
README.md,references/,scripts/,assets/as needed. - Update the relevant section table in the root
README.md. - Bump
versionin both.claude-plugin/*.jsonfiles (minor bump for new skill). - Commit, tag, release.
Adding a new hook
- Create
hooks/<hook-name>/with the hook implementation,install.sh, andREADME.md. - If the hook ships pre-built binaries, add a release flow (see
hooks/balanced-safety-hooks/as the canonical example: cross-platform Go build viamake build-all, GitHub release withSHA256SUMS,install-prebuilt.shthat pins by tag). - Update the Hooks section in the root
README.md. - Bump
versionin both.claude-plugin/*.jsonfiles. - Commit, tag, release.
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.

