publish-plugins
Skill142 starsChanged 24 days agoMITView on GitHub
(project) Use when editing any file under skills/ or plugins/ to bump the plugin version and check the manifests
In this repository14 other files it gives its agents.
- AGENTS.mdAGENTS.md
- upgrade-playbooksSkill
- audit-agent-settingsSkill
- best-practicesSkill
- blindspotSkill
- commitSkill
- differenceSkill
- fuck-over-engineeringSkill
- prSkill
- refactor-agents-mdSkill
- refactor-memorySkill
- refactor-skillSkill
- simple-englishSkill
- write-like-meSkill
The file
We split this skill into the rules it gives an agent
5 rules. About 0 are common, 2 are distinctive, and 3 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.
3
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 42
If a plugin's version already differs from origin/main, it was bumped for unreleased work.
- Very specific: line 50
Commit a bump made on its own through the commit skill with no argument: its subject already says why a bump exists.
- Very specific: line 58
Every plugin is described twice on purpose: once in its own .claude-plugin/plugin.json and once as an entry in …
- Distinctive: line 61
every component path either manifest declares exists on disk.
- Distinctive: line 62
every skill directory in a plugin's root appears in both skill lists.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill 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
Publish Plugins
Prepare the repo's plugins for release: work out which ones changed, bump their versions, and check the manifests still describe what is on disk.
Each plugin owns its version in its own .claude-plugin/plugin.json, and the repo-root .claude-plugin/marketplace.json carries no version fields:
| Plugin | Manifest |
hal-skills | skills/.claude-plugin/plugin.json |
hal-output-styles | plugins/hal-output-styles/.claude-plugin/plugin.json |
hal-grammar-check | plugins/hal-grammar-check/.claude-plugin/plugin.json |
hal-session-auto-rename | plugins/hal-session-auto-rename/.claude-plugin/plugin.json |
hal-shell-mode-auto-suggest | plugins/hal-shell-mode-auto-suggest/.claude-plugin/plugin.json |
hal-voice | plugins/hal-voice/.claude-plugin/plugin.json |
1. Find the changed plugins
git diff --name-only origin/main...skills/** belongs to hal-skills, plugins/<name>/** to that plugin. Only the plugins with changes are candidates for a bump.
2. Check what is already bumped
For each candidate, compare the working-tree version against origin/main:
git show origin/main:skills/.claude-plugin/plugin.jsonIf a plugin's version already differs from origin/main, it was bumped for unreleased work. Do not bump it again. Only bump a plugin whose version still matches origin/main. Check each candidate independently — a change under skills/ says nothing about whether hal-voice needs a bump.
3. Bump
- Patch (0.2.0 -> 0.2.1): bug fixes, config changes, style cleanup.
- Minor (0.2.0 -> 0.3.0): new features, new hooks, new commands.
- Major (0.2.0 -> 1.0.0): breaking changes to hook behavior or config format.
Commit a bump made on its own through the commit skill with no argument: its subject already says why a bump exists.
4. Check the manifests are in sync
python3 .claude/skills/publish-plugins/scripts/check_manifest_sync.pyEvery plugin is described twice on purpose: once in its own .claude-plugin/plugin.json and once as an entry in .claude-plugin/marketplace.json. For each plugin the script checks that
- both describe the same fields with the same values, apart from
versionand the directory listing fieldsicon,documentationUrl,supportUrl,privacyPolicyUrl,termsOfServiceUrl(plugin.json only) andsource,category,tags,strict,defaultEnabled(marketplace entry only — Claude Code ignores them in a plugin.json). - every component path either manifest declares exists on disk.
- every skill directory in a plugin's root appears in both skill lists.
It exits non-zero naming each problem. Fix the manifests it names, then run it again.
5. Check the skill descriptions are in sync
python3 .claude/skills/publish-plugins/scripts/check_skill_descriptions_sync.pyFor every skill under skills/, check that interface.short_description in agents/openai.yaml matches its description in both README.md and skills/README.md exactly. Treat the README descriptions as the source of truth. Keep descriptions as single-line double-quoted strings. Fix missing entries or mismatches, then rerun until the check passes.
6. Validate
claude plugin validate .
claude plugin validate ./skills
claude plugin validate ./plugins/hal-output-styles
claude plugin validate ./plugins/hal-grammar-check
claude plugin validate ./plugins/hal-session-auto-rename
claude plugin validate ./plugins/hal-shell-mode-auto-suggest
claude plugin validate ./plugins/hal-voiceThe first call validates the marketplace manifest; the rest validate each plugin manifest.
7. Run the mod tests
make test-claude-code-modsThis runs claude plugin test on every plugin with a tests/ directory, which is what the claude-code-tests CI job runs too. Fix the failing test or the module, then rerun until every test passes; a command that fails to start is not a pass.
8. Check the skills still group for other agents
npx -y skills@latest add . -lThis lists what npx skills add vinta/hal-9000 will offer, reading the working tree instead of the published repo, and installs nothing. Every skill has to appear under the Hal Skills heading with a count matching the skills list. A skill listed without that heading means the hal-skills marketplace entry lost its name or its skills array — that CLI reads only the two manifests at the repo root, so it never sees skills/.claude-plugin/plugin.json.
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.

