g0ld2k Skills

Reusable Agent Skills for commit messages, pull requests, review closeout, release notes, and work orchestration.

Skill Catalog

Repository Shape

This repository is one self-contained plugin described by the Agent Plugins v1 specification.

• plugin.json is the Agent Plugins v1 manifest for this repository.
• skills/<skill-name>/SKILL.md contains the canonical skill instructions and is discovered directly by plugin clients.
• skills/<skill-name>/references/ contains optional supporting notes.
• skills/<skill-name>/scripts/ contains optional helper scripts used by the skill.
• skills/<skill-name>/agents/openai.yaml contains Codex/OpenAI UI metadata.
• .agents/plugins/marketplace.json and .github/plugin/marketplace.json are thin Codex and GitHub Copilot adapters that point at the repository root.
• schemas/agent-plugins/1.0.0/plugin.schema.json is the pinned Agent Plugins v1 manifest schema used by the validator.

Agent Plugins v1 discovers skills from immediate skills/<name>/SKILL.md directories; there is no recursive search and the portable plugin.json cannot list or relocate them. There is no generated copy of skills/: the root manifest and the canonical skill tree are the distributable artifact.

The manifest schema sets additionalProperties: false, so extensions — keyed by reverse-domain namespace — is the only legal home for future plugin-level client-specific data.

Per-skill client behavior lives in that client's own metadata, and an explicit-only skill needs a guard for each install path because neither client reads the other's:

The validator requires both for every skill in EXPLICIT_ONLY_SKILLS, and forbids the Claude field on every other skill. disable-model-invocation is a Claude Code extension rather than a portable Agent Skills field; it is carried deliberately so gh skill install … --agent claude-code cannot leave a state-changing orchestrator implicitly invocable.

Why references/conventions.md is vendored per skill

_shared/conventions.md is the single source of truth, copied into each consuming skill by scripts/sync-shared-conventions.py and drift-checked in CI. The duplication is deliberate: it keeps a single-skill gh skill install … skills/<name>/SKILL.md self-contained, since that installs one skill directory rather than the whole plugin. The sync script discovers consumers from the skills that reference references/conventions.md, so adding or removing that link is all it takes to opt in or out.

Add a New Skill

1. Use docs/skill-template.md as the blueprint, not a literal copy: create skills/<name>/SKILL.md starting from the template's frontmatter block (filled in), then write each section the template prescribes (its quoted ## … headings become your real headings). No template prose survives into the finished skill — delete the DOCS-ONLY blocks, the ## Frontmatter rules section, and every guidance line as you replace it.
2. Create skills/<name>/agents/openai.yaml from the template's stub.
3. Create skills/<name>/references/validation-scenarios.md with at least 3 scenarios (happy path, edge case, adversarial) — the template's Validation Scenarios section points at it.
4. Add a row for the skill to the ## Skill Catalog table above.
5. Run the sync and validation commands:

python3 scripts/sync-shared-conventions.py
python3 scripts/validate-skills-repo.py

Direct Agent Skills Install

Install a canonical skill directly from the repository with gh skill:

gh skill install g0ld2k/Skills skills/commit-message/SKILL.md --agent codex --scope user
gh skill install g0ld2k/Skills skills/pr-generator/SKILL.md --agent claude-code --scope user
gh skill install g0ld2k/Skills skills/pr-comment-review/SKILL.md --agent github-copilot --scope project

Use exact skills/<skill-name>/SKILL.md paths to install one skill instead of the complete plugin.

Validate the direct publisher shape:

gh skill publish --dry-run

Marketplace Install Examples

Codex:

codex plugin marketplace add g0ld2k/Skills --ref main
codex plugin add g0ld2k-skills@g0ld2k-skills

GitHub Copilot CLI:

copilot plugin marketplace add g0ld2k/Skills
copilot plugin install g0ld2k-skills@g0ld2k-skills

Validation

Run the same structural checks used by CI:

python3 scripts/validate-skills-repo.py
python3 scripts/sync-shared-conventions.py
git diff --exit-code
status="$(git status --porcelain)"
if [ -n "$status" ]; then echo "$status"; exit 1; fi
python3 -m unittest discover -s tests
find plugin.json .agents .github schemas -name '*.json' -print |
  while IFS= read -r file; do python3 -m json.tool "$file" >/dev/null; done
find skills scripts -type f -name '*.sh' -print |
  while IFS= read -r file; do bash -n "$file"; done
gh skill publish --dry-run

The validator checks the Agent Plugins v1 manifest against the pinned schema, skill frontmatter and naming, agents/openai.yaml metadata, cross-skill references, local links, and shared-conventions freshness. CI additionally checks JSON syntax, shell script syntax, and the gh skill publisher shape. It does not run client runtime validation or skill evals.

Optional local checks:

codex plugin marketplace add ./.
codex plugin add g0ld2k-skills@g0ld2k-skills

Security

Skills can influence agent behavior and may guide state-changing actions such as commits, pull request updates, comments, or merges. Inspect third-party skills before installing them, and keep explicit approval gates for write operations.