Skills Repo

Canonical source for reusable skills that can be developed once under skills/ and then consumed by Codex, GitHub Copilot, and future AI CLI tools.

What This Repo Contains

• Canonical skills under skills/<skill-name>/
• Generated OpenAI/Codex-facing metadata under each skill's agents/
• Generated compatibility wrappers under compat/
• Repo-level validation, metadata sync, and smoke-test tooling under scripts/
• Tests and fixtures under tests/

Repo Structure

.
├── README.md
├── Makefile
├── AGENTS.md
├── .github/copilot-instructions.md
├── docs/
│   └── skill-platform-playbook.md
├── scripts/
├── skills/
│   ├── INDEX.md
│   └── <skill-name>/
├── compat/
│   ├── copilot/
│   └── claude-code/
└── tests/

Key Files

• docs/skill-platform-playbook.md: Canonical workflow for creating, updating, testing, and packaging skills.
• AGENTS.md: Codex-oriented repo instructions.
• .github/copilot-instructions.md: GitHub Copilot-oriented repo instructions.
• skills/INDEX.md: Generated index of available skills and local/distributed usage patterns.
• manifest.json: Generated repo manifest for distribution/discovery metadata.
• Makefile: Human-friendly entrypoints for the common workflows.

Current Skills

repo-oversight-evaluation

Path: skills/repo-oversight-evaluation/SKILL.md

Purpose:

• Evaluate a repository or pull request for AI-assisted delivery, human oversight risk, and security review priority.
• Produce a comparable, evidence-based report for AppSec and R&D triage.
• Support PR-scoped review and follow-up cleanup/test work.

Deep dive:

• Canonical skill: skills/repo-oversight-evaluation/SKILL.md
• Report contract: skills/repo-oversight-evaluation/references/report-contract.md
• PR mode guidance: skills/repo-oversight-evaluation/references/pr-mode.md
• Deterministic evaluator: skills/repo-oversight-evaluation/scripts/evaluate_repo_oversight.py
• Generated OpenAI metadata: skills/repo-oversight-evaluation/agents/openai.yaml
• Generated Copilot wrapper: compat/copilot/repo-oversight-evaluation.md
• Generated Claude Code wrapper: compat/claude-code/repo-oversight-evaluation.md

Common Workflows

Use npm for the standard cross-platform repo workflows:

npm run sync:metadata
npm run build:catalog
npm run validate
npm test
npm run test:skill -- repo-oversight-evaluation

Optional Unix-like convenience commands are also available through the Makefile:

make help
make sync
make build
make validate
make test
make test repo-oversight-evaluation

Equivalent direct commands through the shared Python launcher:

node scripts/run_python.mjs scripts/sync_metadata.py
node scripts/run_python.mjs scripts/build_catalog.py
node scripts/run_python.mjs scripts/validate_repo.py
node scripts/run_python.mjs scripts/test_repo.py
node scripts/run_python.mjs scripts/test_skill.py repo-oversight-evaluation

Smoke-test artifacts and evaluator outputs are written under test_results/, which is git-ignored.

Local Testing Without Installation

Use a skill directly from this checkout:

Use $<skill-name> at ./skills/<skill-name> ...

If a runner requires an absolute path, resolve it at runtime from the repo root.

Default evaluator reports go to test_results/.

Distribution

Publish the repo and install/update a skill by repo URL:

npx skills add <repo-url> --skill <skill-name>
npx skills update <repo-url> --skill <skill-name>

Where To Start

• If you want to create or update skills, start with docs/skill-platform-playbook.md.
• If you want to browse what exists, start with skills/INDEX.md.
• If you want the day-to-day commands, use Makefile.