Skill Management
Overview
Use this skill to turn useful local skill workflows into public, reusable, and testable skill packages. For publish-before-review privacy and contract checks, use publication-review/.
Workflow
- Identify the task type.
- Use
skill-management/ for inventory, audit, standards, tests, migration, and publishing workflows.
- Use
publication-review/ for final public-safety, structure, and truthful-contract checks before publication.
- Use
skill-examples/<skill-name>/ for mature reusable skills and public-safe examples.
- Inspect the candidate skill or workflow.
- Read
SKILL.md, agents/openai.yaml, and directly referenced resources.
- Search for private paths, user identifiers, unpublished content, account details, and local-only assumptions.
- Normalize the public contract.
- Keep only implemented behavior in the trigger description and body.
- Mark unsupported behavior as manual, optional, or out of scope.
- Replace private examples with generic fixtures.
- Validate structure and metadata.
- Ensure the folder name is lowercase hyphen-case.
- Ensure frontmatter follows the system
skill-creator validator; name and description are required.
- Ensure optional resources are actually used.
- Ensure reusable
skill-examples/ skills include install.md with copy,
dependencies, install, update, verification, rollback, and notes sections.
- Ensure required companion skills, CLIs, app connectors, and bundled scripts
are explicit enough for a post-install agent check.
- Ensure installed path examples resolve from
$CODEX_HOME/skills, with
~/.codex/skills as the fallback when CODEX_HOME is unset.
- Record the result.
- Update collection-level documentation when a new category or convention is introduced.
- Avoid adding per-skill README files unless explicitly required by the repository owner.
Standard Skill Shape
A public skill should normally look like:
skill-name/
SKILL.md
install.md
agents/openai.yaml
references/
scripts/
assets/
Only keep optional directories that serve the skill. Delete scaffold placeholders.
Validation Commands
When the system skill-creator utilities are available, validate a skill with:
uv run --with pyyaml python .\skill-examples\skill-creator\scripts\quick_validate.py .\skill-management
If the repository sample is unavailable, use an installed skill-creator validator without committing machine-specific absolute paths.
For skill examples, also run the installation smoke checker in source mode:
python .\scripts\validate_muxing_install.py --source-only --skill deepseek-agent
Output Expectations
- For audits, report blocking publication issues first.
- For migrations, list changed files and any behavior intentionally omitted.
- For publication-readiness validation, delegate to
publication-review/ and distinguish structural validation from real-world forward testing.
- For public README updates, keep the repository purpose broad and avoid private context.
1---2name: skill-management3description: Manage, inventory, standardize, migrate, and test reusable Codex skills for the MetaSkill repository. Use when Codex needs to inspect local skills, convert a personal workflow into a public skill candidate, normalize skill folder shape, plan skill promotion, or decide whether a skill belongs in skill-management, publication-review, or skill-examples.4---56# Skill Management78## Overview910Use this skill to turn useful local skill workflows into public, reusable, and testable skill packages. For publish-before-review privacy and contract checks, use `publication-review/`.1112## Workflow13141. Identify the task type.15 - Use `skill-management/` for inventory, audit, standards, tests, migration, and publishing workflows.16 - Use `publication-review/` for final public-safety, structure, and truthful-contract checks before publication.17 - Use `skill-examples/<skill-name>/` for mature reusable skills and public-safe examples.182. Inspect the candidate skill or workflow.19 - Read `SKILL.md`, `agents/openai.yaml`, and directly referenced resources.20 - Search for private paths, user identifiers, unpublished content, account details, and local-only assumptions.213. Normalize the public contract.22 - Keep only implemented behavior in the trigger description and body.23 - Mark unsupported behavior as manual, optional, or out of scope.24 - Replace private examples with generic fixtures.254. Validate structure and metadata.26 - Ensure the folder name is lowercase hyphen-case.27 - Ensure frontmatter follows the system `skill-creator` validator; `name` and `description` are required.28 - Ensure optional resources are actually used.29 - Ensure reusable `skill-examples/` skills include `install.md` with copy,30 dependencies, install, update, verification, rollback, and notes sections.31 - Ensure required companion skills, CLIs, app connectors, and bundled scripts32 are explicit enough for a post-install agent check.33 - Ensure installed path examples resolve from `$CODEX_HOME/skills`, with34 `~/.codex/skills` as the fallback when `CODEX_HOME` is unset.355. Record the result.36 - Update collection-level documentation when a new category or convention is introduced.37 - Avoid adding per-skill README files unless explicitly required by the repository owner.3839## Standard Skill Shape4041A public skill should normally look like:4243```text44skill-name/45 SKILL.md46 install.md47 agents/openai.yaml48 references/49 scripts/50 assets/51```5253Only keep optional directories that serve the skill. Delete scaffold placeholders.5455## Validation Commands5657When the system skill-creator utilities are available, validate a skill with:5859```powershell60uv run --with pyyaml python .\skill-examples\skill-creator\scripts\quick_validate.py .\skill-management61```6263If the repository sample is unavailable, use an installed `skill-creator` validator without committing machine-specific absolute paths.6465For skill examples, also run the installation smoke checker in source mode:6667```powershell68python .\scripts\validate_muxing_install.py --source-only --skill deepseek-agent69```7071## Output Expectations7273- For audits, report blocking publication issues first.74- For migrations, list changed files and any behavior intentionally omitted.75- For publication-readiness validation, delegate to `publication-review/` and distinguish structural validation from real-world forward testing.76- For public README updates, keep the repository purpose broad and avoid private context.