adviselisted

Diagnose a project's test gaps and name the cheapest next tests. Use after hs-setup, after plugin install, or when someone asks what to test, which framework to add, whether coverage is enough, or to audit the test suite. Triggers: "/advise", "what should we test", "which test framework", "is coverage enough", "audit this project's tests", "quality report".
haru3613/harness-ship · ★ 1 · Testing & QA · score 67

Install: claude install-skill haru3613/harness-ship

# advise Speak as a senior test engineer. Inspect the repository, then name at most three **next cuts**. Write `.harness-ship/quality-report.md` from [quality-report-template.md](quality-report-template.md). This skill does not draft a Test Contract, install a tool, or return a release verdict. Read repository instructions and the `## harness-ship` Config v3 block first. Run `hs-setup` only when the block is absent or unsupported. ## 1 — Inventory Collect what the project already has: - product surface and the journeys a user can finish; - the complete test tree, runners, and CI jobs, including path-gated no-ops; - red, flaky, skipped, and empty suites, each named; - defect history that clusters by failure mechanism, not by file; and - any existing `.harness-ship/test-contract.md` or `.harness-ship/quality-report.md`. An empty repository is a valid inventory: no runner, no CI, no journeys yet. Stop inventory when another file would not change the next cut. ## 2 — Diagnose the portfolio Name the **shape** from the inventory: - **pyramid** — many fast narrow tests, a few integration seams, almost no UI E2E; - **ice-cream** — most confidence sits in slow UI or device tests; - **hourglass** — unit and E2E exist, the wiring in the middle does not; - **empty** — no automated proof of user-visible behaviour; - **thin-middle** — parsers or widgets are green, the content script, handler, or client that joins them is unproven. Use coverage as a **flashlight**: list risky mo