Agent Skills

Reusable Claude Code skills for software engineering agents.

Structure

Each skill lives in its own directory with a SKILL.md (practice guide) and supporting tools.

• event-modeling/ — event modeling skill with Pydantic schema, validation, and SVG rendering
• event-modeling/ts/ — the model as typed TypeScript; assembles to the JSON, generates the protobuf, em CLI
• event-modeling/ts/viewer/ — the live diagram em view serves; Vue over the assembled JSON, built by Vite into viewer/dist
• .claude/skills/lazy/ — response contract skill plus its Stop hook, installed active for this repo

The lazy skill is edited here, nowhere else

~/.claude/skills/lazy/ and ~/.codex/skills/lazy/ are symlinks into .claude/skills/lazy/, so this repo is the only copy. Edit it here, open a PR, and after merge publish it so teammates get the same version:

gns skills publish .claude/skills/lazy -m "<what changed>" --team architecture --public --force
gns skills repair --key user.noah.skills.lazy

Editing ~/.claude/skills/lazy/ directly writes through the symlink into an unreviewed working copy of this repo. Commit it.

Dependencies

pip install pydantic pytest

Running Tests

pytest event-modeling/tests/ -v
pytest tests/ -v                   # lazy-check.py
cd event-modeling/ts && npm test   # tsc, biome, vite build, node --test, buf lint

Test Harness

Three kinds:

1. Unit tests (tests/test_event_model.py) — the Python tool's parsing, validation and rendering
2. TypeScript tests (ts/test/) — typecheck.ts holds one @ts-expect-error per compiler check, so tsc fails if a check stops firing; the rest cover assembly, the proto generator, the viewer's layout (pure, no browser) and the CLI, including the view server. Biome does not format typecheck.ts: the directive binds to the next line, so reformatting a chain moves the error off it and silently disarms the check
3. Brief coverage (ts/test/brief-coverage.test.ts) — whether the model answers the product brief, which is the only thing left that no type system can decide

Test Fixtures

• Product brief (tests/fixtures/todo_app_brief.md) — a plain-text product description, the real input an agent works from
• The model (ts/examples/todo-app/) — the model built from that brief. ts/test/todo-app.json is what it assembles to, and event_model.py validate reports zero warnings on it
• Golden JSON (tests/fixtures/todo_app_model.json) — a hand-written model of the same brief, for the Python tests

Conventions

• Skills use SKILL.md frontmatter with name and description
• Tool scripts are self-contained Python with minimal dependencies
• No coauthor signatures in commits