maintain-skills
Skill159 starsChanged 10 days agoMITView on GitHub
Maintain Agent Skills in a Heptabase repository across Claude Code, Codex, and Cursor. Use when creating, editing, moving, removing, or auditing files under .claude/skills/.
6 other projects use a version of this.
Adopted by
Most recently taken up by
- vic-lsh/agent-skillstheir own version, first seen Nov 8, 2026
- christofferbergj/agent-skillstheir own version, first seen Oct 15, 2026
- Neplich/h-level-model-skillstheir own version, first seen Sep 14, 2026
- Neplich/dev-agent-skillstheir own version, first seen Sep 14, 2026
- ltatarev/skillstheir own version, first seen Aug 9, 2026
- mamachanko/skillstheir own version, first seen Jul 11, 2026
Other versions6 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 6
- Neplich/dev-agent-skills★ 8same name
- vic-lsh/agent-skills★ 2same name
- christofferbergj/agent-skills★ 1same name
- mamachanko/skills★ 1same name
- ltatarev/skills★ 0same name
- Neplich/h-level-model-skills★ 0same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository5 other files it gives its agents.
- AGENTS.mdAGENTS.md
- CLAUDE.mdCLAUDE.md
- public-repo-guardSkill
- releaseSkill
- heptabase-cliSkill
The file
Show its rules and text
We split this skill into the rules it gives an agent
24 rules. 0 are common, 15 are distinctive, and 9 are very specific to this project.
0
Common
Most of the 6 edited versions we read keep these.
15
Distinctive
This author's own choices. Worth a look.
9
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Very specific: line 15
Treat .claude/skills/<skill-name>/ as the canonical source for team-owned repository skills.
- Very specific: line 16
Keep .agents/skills as a symlink to ../.claude/skills so Codex and other Agent Skills clients discover the same skills.
- Very specific: line 17
Do not copy team skills into .cursor/skills/ or .codex/skills/.
- Very specific: line 18
Use convert-cursor-to-claude-skill when converting rules, migrating commands, or fixing misplaced skill directories.
- Distinctive: line 36
When the user has not chosen:
- Distinctive: line 38
Prefer model invocation for repository conventions, reference knowledge, and guidance that should apply whenever relevant.
- Distinctive: line 39
Consider explicit-only invocation for named workflows whose timing the user is expected to control.
- Distinctive: line 40
Do not infer explicit-only invocation only because a workflow uses tools, requires approval, or can cause side effects.
- Distinctive: line 41
If the correct invocation behavior is not easy to judge from the skill's purpose, ask the user before finalizing the skill.
- Distinctive: line 49
Follow the user's request for Codex metadata.
- Very specific: line 51
Do not add the file only because a skill is new or being edited.
- Distinctive: line 62
When adding or editing the file:
- Very specific: line 66
When present or required, interface.short_description must contain 25 to 64 characters, and interface.default_prompt must mention …
- Distinctive: line 70
Run the repository checker after creating or changing a skill:
- Distinctive: line 78
Before finalizing a new or changed skill:
- Very specific: line 80
Run check-skills.mjs --changed to detect mechanical conflicts.
- Very specific: line 81
Search .claude/skills/ and relevant AGENTS.md, CLAUDE.md, .claude/rules/, and .cursor/rules/ files for overlapping triggers or instructions.
- Distinctive: line 86
Keep copies of the same team skill aligned across repositories.
- Distinctive: line 87
If the correct behavior is unclear, ask the user.
- Distinctive: line 91
Run the checker from the repository root:
- Distinctive: line 111
When a task changes skills in more than one repository, run the checker separately in each affected repository.
- Very specific: line 115
Only canonical .claude/skills/ files changed.
- Distinctive: line 117
No affected skills give conflicting instructions.
- Distinctive: line 118
No unrelated skill or agent configuration was changed.
“Common” is measured: kept by at least 6 in 10 of the 6 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
Maintain Skills
Apply only the Heptabase repository conventions below. Follow the active agent's normal skill-authoring guidance for general skill structure and writing quality.
Use the canonical locations
- Treat
.claude/skills/<skill-name>/as the canonical source for team-owned repository skills. - Keep
.agents/skillsas a symlink to../.claude/skillsso Codex and other Agent Skills clients discover the same skills. - Do not copy team skills into
.cursor/skills/or.codex/skills/. - Use
convert-cursor-to-claude-skillwhen converting rules, migrating commands, or fixing misplaced skill directories. This skill owns skill content and metadata, not path migration.
Decide the invocation policy
For every new skill, explicitly decide whether the model may invoke it automatically.
For an existing skill, preserve its current invocation policy unless the user asks to change it or the skill's intended behavior has materially changed.
The user's stated invocation choice always wins; you can suggest or ask but not override it. Do not override it with the defaults below.
<!-- prettier-ignore -->| Intended behavior | SKILL.md | agents/openai.yaml |
| The agent may load it when relevant | Omit disable-model-invocation, or set it to false | Omit policy.allow_implicit_invocation, or set it to true |
| Only the user may start it | Set disable-model-invocation: true | Set policy.allow_implicit_invocation: false |
Because Codex allows implicit invocation by default, a human-invocation-only skill must include both settings and the complete UI metadata described below. An implicitly invocable skill may omit agents/openai.yaml unless it needs Codex UI metadata or dependencies.
When the user has not chosen:
- Prefer model invocation for repository conventions, reference knowledge, and guidance that should apply whenever relevant.
- Consider explicit-only invocation for named workflows whose timing the user is expected to control.
- Do not infer explicit-only invocation only because a workflow uses tools, requires approval, or can cause side effects. Authorization is still checked when the action occurs.
- If the correct invocation behavior is not easy to judge from the skill's purpose, ask the user before finalizing the skill.
Add Codex metadata by invocation mode
For a skill that the model may invoke automatically, agents/openai.yaml is optional. Add or update it only when it serves a Codex-specific purpose:
- Improve a user-facing skill's name, description, default prompt, icon, or branding in Codex.
- Declare tool dependencies for Codex.
- Follow the user's request for Codex metadata.
Do not add the file only because a skill is new or being edited. Keep portable discovery in the SKILL.md name and description.
For a human-invocation-only skill, agents/openai.yaml and these fields are required:
interface.display_nameinterface.short_descriptioninterface.default_promptpolicy.allow_implicit_invocation: false
The skill selector is the primary entry point for a human-invocation-only skill, so its name, description, and starter prompt must be clear there.
When adding or editing the file:
- Quote string values.
- For implicitly invocable skills, keep interface fields optional.
- When present or required,
interface.short_descriptionmust contain 25 to 64 characters, andinterface.default_promptmust mention$skill-name. - Include
policy.allow_implicit_invocationwhen needed to keep Codex behavior aligned withdisable-model-invocation. - Preserve unrelated interface, policy, and dependency fields.
Run the repository checker after creating or changing a skill:
node .claude/skills/maintain-skills/scripts/check-skills.mjs --changedPrevent conflicting skills
Before finalizing a new or changed skill:
- Run
check-skills.mjs --changedto detect mechanical conflicts. - Search
.claude/skills/and relevantAGENTS.md,CLAUDE.md,.claude/rules/, and.cursor/rules/files for overlapping triggers or instructions. Read possible overlaps; the checker cannot judge whether prose is consistent. - Resolve each conflict introduced or affected by the change:
- Frontmatter names must be unique within one repository. Separate repositories may reuse names.
- For overlapping triggers, narrow the descriptions and state which skill owns each case.
- For conflicting instructions, keep one source of truth or narrow the scopes so both can coexist.
- Keep copies of the same team skill aligned across repositories.
- If the correct behavior is unclear, ask the user. Report unrelated pre-existing conflicts without expanding the task to fix them.
Validate the result
Run the checker from the repository root:
node .claude/skills/maintain-skills/scripts/check-skills.mjs --changedFor a full repository audit:
node .claude/skills/maintain-skills/scripts/check-skills.mjsStandard checks require complete Codex UI metadata only for human-invocation-only skills.
For an optional audit that requires complete Codex UI metadata for every existing skill:
node .claude/skills/maintain-skills/scripts/check-skills.mjs --strict-uiWhen a task changes skills in more than one repository, run the checker separately in each affected repository.
Review the final diff and confirm that:
- Only canonical
.claude/skills/files changed. - Invocation settings agree across
SKILL.mdandagents/openai.yaml. - No affected skills give conflicting instructions.
- No unrelated skill or agent configuration was changed.
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.

