skill-unit

Skill Unit

A plugin that brings structured, reproducible unit testing to AI agent skills.

What it does

Skill Unit lets you write test specs for AI agent skills using a familiar unit-testing mental model — define prompts, declare expected outcomes, and get pass/fail results. It uses process-level isolation to ensure unbiased evaluation: each test prompt runs in a separate CLI session that has no access to expectations or any indication it is being tested.

Key features

• Spec files (*.spec.md) — test cases written as prompts with expectations, grouped into suites with YAML frontmatter
• Anti-bias execution — each test prompt runs in an isolated CLI process with no access to expectations, test metadata, or the test directory
• Harness-agnostic — configurable CLI runner works with any AI agent harness (Claude Code, Copilot, Codex, etc.)
• Checked-in results — timestamped results files commit to your repo for regression tracking via git history
• Fixtures & setup/teardown — declare filesystem state and run polyglot scripts before/after tests
• CI/CD ready — run headless with your agent harness of choice

Quick start

1. Install the plugin in your project
2. Create a skill-tests/ directory with *.spec.md files (see skills/skill-unit/templates/example.spec.md)
3. Run /skill-unit or ask your agent to "run skill tests"

Status

Phase 1 (MVP) — in development.