Skill Integration Tester parses the Multi-Skill Workflows section of a project’s CLAUDE.md, resolves each step to a skill directory and checks that the output of step N fits the input of step N+1. It also reports file naming violations and marks every workflow as valid, broken or warning.
A dry-run mode writes synthetic fixture JSON files so contracts can be checked without real data. It runs offline with the standard library, needs no API keys and only reads SKILL.md and CLAUDE.md files.
What You Get
- A JSON report with totals of valid, broken and warning workflows
- A Markdown report with step checks, handoff results (PASS, FAIL or N/A) and naming violations
- Synthetic fixture files under reports/fixtures when run in dry-run mode
- A contract reference describing handoff patterns between skills
Common Use Cases
- Verifying pipeline health before releasing new skills
- Debugging a broken handoff after changing a skill’s output format
- Running as a pre-check in CI for pull requests that touch skill scripts
What Makes It Unique
It treats the workflow list in CLAUDE.md as the specification and tests the handoffs between skills deterministically, so the same inputs always give the same verdict. It never modifies any skill.
Under The Hood
A standard-library Python script reads CLAUDE.md and the SKILL.md files of the skills it resolves, with no network access. It writes timestamped JSON and Markdown reports, and in dry-run mode fixture JSON files, into a reports folder. The SKILL.md has the model run the script from the shell.