This repository contains reusable agent tooling. It currently focuses on agent skills, with room for other agentic work over time.
Use pnpm for all commands.
- Source folders live under
src/. - Each skill should have its own folder, named with lowercase kebab-case.
- Each skill folder should include a
SKILL.mdentrypoint withnameanddescriptionYAML frontmatter. - Install local skills by symlinking skills from
src/skills/into.agents/skills. - Keep repo tests outside published skill folders. Put skill tests under
tests/skills/, not undersrc/skills/<skill-name>/, so installed or packaged skills do not include test files.
Run focused validation on what you changed.
- Rewrite TypeScript, JavaScript, JSON, and other non-Markdown files with
pnpm exec biome check --write <file-or-path>. - Verify with
pnpm exec biome ci <file-or-path>before finishing. - Python uses standard-library formatting conventions in this repo for now. Validate Python changes
with
python3 -B -m unittest discover -s testsorpnpm test; pre-commit only compiles staged Python files. - Markdown and MDX use Prettier with
.git-hooks/prettierrc.markdown.jsononly. Do not add a repo-wide.prettierrc. - For agent-authored Markdown (
*.md,*.mdx), wrap prose at roughly 80 columns. This includes skill docs, automation docs, prompt docs, and YAML frontmatter descriptions. Do not reflow tables, code blocks, URLs, or generated inventory when wrapping would hurt readability. - Lefthook pre-commit runs Biome and Prettier on staged files when installed. Cursor and Claude editor hooks use the same Markdown config.
Run these when they apply to your change:
pnpm format:check
pnpm tc
pnpm testpnpm check runs format check, typecheck, and tests together.