haorantang97/LabKit / skills/skill-skill
skill-skill
VerifiedCreate, revise, validate, package, or publish AI skills and rule files. Use as the main entry for the user's skill-authoring workflow, including creating a new SKILL.md, improving an existing rule, and preparing a skill repository. Choose only the requested stage; ordinary product code or content work is outside this skill.
| File | Kind | Size |
|---|---|---|
| LICENSE | other | 4.5 KB |
| SKILL.md | skill md | 3.9 KB |
| references/authoring.md | markdown | 4.6 KB |
| references/publishing.md | markdown | 4.7 KB |
| references/repository-presentation.md | markdown | 8.1 KB |
references/authoring.md
View on GitHub# Authoring and validation
Read this for creating a skill, changing its behavior or structure, validating it, or selecting an official helper. Packaging-only work may use the existing modules directly.
## Start with the capability
Define what a successful invocation produces and which requests need the skill. Use examples from the task or observed failures. Ask for more only when the missing distinction would materially change the design.
Assume the model is already capable. Add instructions that change decisions or improve reliability. Preserve scope, chosen artifacts and version; do not convert a local edit into a publication, review pipeline or broad configuration audit.
Keep discovery concise and specific. Required frontmatter is name and description; preserve supported existing metadata. Use a lowercase letters/digits/hyphens name under 64 characters. Other rule formats retain their own host requirements rather than receiving SKILL frontmatter indiscriminately.
## Choose resources by need
- Keep shared purpose, real constraints and useful routing in SKILL.md.
- Use references for substantial detail relevant only to particular modes, and link them at the point of use.
- Use scripts for repeated or fragile deterministic operations; verify changed scripts by running them safely.
- Use assets for files that belong in the generated output. Repository covers and their design or generation records belong outside the installed skill; classify files by their role in the workflow, not their extension or directory name.
- Do not create empty resource directories, placeholder examples, a README, a changelog or an extra router without a task reason. Publishing a repository can provide that reason; a small local skill often does not.
There is no universal minimum description length, mandatory five-part trigger, numeric constraint quota, fixed test count or minimum size before moving detail into references. Choose specificity and structure for the actual risk and workflow.
## Official tools remain the host-specific reference
In Codex, resolve the current official skill-creator from the live skill catalog. A standard location is the active CODEX_HOME's skills/.system/skill-creator directory. Read its relevant SKILL.md and helper usage before relying on current flags or supported metadata. On other hosts or when absent, use that host's documented format and state which checks were unavailable.
Typical official helpers, run only when appropriate:
```text
python3 <official-skill-creator>/scripts/init_skill.py <name> --path <destination>
python3 <official-skill-creator>/scripts/quick_validate.py <skill-folder>
```
Use the initializer for a new skill when it helps; do not reinitialize an existing skill. Request only resource folders the task needs. For agents/openai.yaml, read the current official metadata reference. A generator can replace the whole YAML file: preserve policy, dependencies and unrelated fields when editing existing metadata. Preserve automatic discovery unless the user requested a different invocation policy.
For installation, use the current official skill-installer when applicable. Install the intended version, preserve customizations and inspect the resulting files in the correct runtime. For packaging or installation changes, follow the [installation boundary and verification guidance](publishing.md#installation-boundary). Registration is not evidence of host loading, Hook trust or actual delivery.
Keep official system skills and helper scripts intact and updateable. Do not ship copies of host scripts or assume this machine's absolute paths are portable. If official guidance changes, compare the relevant change with this guide before applying it; user conventions and host constraints still govern the task.
## Validate behavior that matters
Run the official validator when available. Structural checks do not establish that the skill makes good decisions. Inspect discovery scope, reference reachability, permission boundaries and the actual change.
For a substantial revision, use realistic cases from the capability. Include a narrow task that should stay narrow and an out-of-scope request when misrouting is a concern. Use independent evaluation only when complexity or risk justifies it and host and user permit delegation. Do not turn every small edit into a mandatory evaluation pipeline.
Testing a local authoring workflow must not publish a repository, send a message, delete user data or grant Hook trust. Those steps require corresponding task authorization. Distinguish structural validation, a reasoned walkthrough, an executed example and actual client behavior in the receipt.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |