codex-skills

A curated collection of Codex skills for recurring code.neureka.dev workflows, rewritten from the companion Claude command set into the recommended Codex skill format.

Install

Place skill folders from skills/ into one of these locations:

• Repo-local: $REPO_ROOT/.agents/skills/<skill-name>
• User-global: $HOME/.agents/skills/<skill-name>

Skills

conventional-commits

Creates and refines commit messages that comply with Conventional Commits v1.0.0, auto-stages the full worktree when nothing is already staged, skips test when the workspace has no tests, and finishes the commit flow by pushing the current branch to origin with Git LFS checks when needed.

Triggering requests:

• Requests to create a commit.
• Requests to write or edit a commit message.
• Requests to run git commit.

Core behavior:

• Inspect staged changes before drafting the message.
• If nothing is staged, inspect the worktree and stage everything with git add -A; if the repo is already clean, stop because there is nothing to commit.
• Run safety checks for risky staged paths and obvious secrets in staged diffs.
• Require applicable lint, test, and build checks before committing.
• Skip check classes that are clearly not needed for the staged change set.
• Skip test when the workspace does not appear to contain any tests and no reliable test command is discoverable.
• Reuse same-session successful lint, test, or build results when no newer relevant code, config, or dependency changes occurred afterward.
• Honor explicit user instructions to skip named classes such as lint, test, or build.
• Fix lint findings, test failures, and code-related build warnings when reasonably possible, rerun only the affected classes, and stop if failures or actionable warnings remain.
• After a successful commit, resolve the current branch and push it to origin.
• Check for Git LFS-tracked content before pushing, and stop if LFS is required but unavailable or blocked.
• Enforce Conventional Commits header format: <type>[optional scope][!]: <description>.
• Require BREAKING CHANGE: footers when applicable.
• Return commit messages that are directly usable.

forgejo-workflows

Creates, updates, and troubleshoots Forgejo Actions workflow files for CI/CD pipelines.

Triggering requests:

• Requests explicitly about Forgejo workflows or Forgejo actions.
• If a request says only "workflow" or "actions", ask whether the user means GitHub or Forgejo before proceeding.

Core behavior:

• Write workflow files under .forgejo/workflows/.
• Use https://forgejo.org/docs/next/user/actions/reference/ as the syntax and behavior reference.
• Use https://code.neureka.dev/api/v1/repos/search?team_id=3&sort=alpha&limit=1000 to discover valid actions.
• Build action identifiers from JSON as {owner.login}/{name}, then use uses: as {owner.login}/{name}@{ref} (example: actions/docker-build-push@v6).
• For tag-based action refs, pin to major version if possible, then major.minor, then the latest version tag; use latest only as fallback.
• Use https://code.neureka.dev/api/v1/repos/search?team_id=4&sort=alpha&limit=1000 to build valid runs-on values from {name}-{version} when available (otherwise use name), preferring latest when present.
• Use https://code.neureka.dev/api/v1/repos/<owner>/<name>/tags (example: /api/v1/repos/actions/docker-build-push/tags) to validate tag-based action refs.
• Prefer validated uses: actions over shell when a suitable action exists.
• Treat sha256 values as invalid for runs-on (runner labels are manually mapped in Forgejo Runner config); sha256 is valid for action refs.