Teach your agent how to build Agent Skills properly — the directory structure, how to write a SKILL.md that triggers at the right time, how to design scripts agents can actually use, and how to handle dependencies so consumers never run an install step.
You're building skills with TypeScript and Bun. Maybe you've written one before and it worked, but the description was too vague (triggered on the wrong prompts), the scripts mixed diagnostics into stdout (broke agent parsing), or the structure didn't survive npx skills add (files outside the skill directory got left behind).
This skill encodes the conventions that avoid those problems.
| Area | What it covers |
|---|---|
| SKILL.md authoring | Frontmatter schema, description writing (trigger keywords, negative scope), body structure, progressive disclosure, Claude Code extensions |
| Script design | JSON to stdout, diagnostics to stderr, --help convention, error messages that help agents self-correct, output size limits, idempotency |
| Directory structure | scripts/ as public interface, src/ for implementation, shell wrapper pattern, what goes at skill root vs inside src/ |
| Dependencies | Lockfile pinning, Bun auto-install, why node_modules must be gitignored, inline version specifiers for simple scripts |
| Testing | Bun test patterns, mocking, factory functions, no live API calls |
| Repo layout | Why skills go in skills/, how the directory boundary keeps dev infrastructure out of installed artifacts |
npx skills add TimBeyer/agent-skills --skill skill-devNo dependencies. Includes a full project spec as a reference document for CI templates, compiled binaries, shared code patterns, and other details your agent can look up when needed.
MIT