| created | "2026-07-12T00:00:00.000Z" |
| modified | "2026-07-12T00:00:00.000Z" |
| reviewed | "2026-07-12T00:00:00.000Z" |
| allowed-tools | Task, Read, Glob, Grep, Bash(git log *), Bash(git diff *), Bash(rg *), TodoWrite |
| args | [path|crate] [--models <a,b>] |
| argument-hint | [path or crate to review] [--models glm-5.2,gemini-2.5-pro] |
| description | Evaluate whether a test suite actually tests what it aims to, using an adversarial pass plus a tests-hidden cold read, then verify every finding against the code. Use when a suite is green but you doubt it would catch a real regression. |
| name | test-strategy-review |
Test Strategy Review
A green suite is not evidence that the tests test what you're aiming for. It is
evidence that the tests you wrote pass. The gap between those two is where the
expensive bugs live — and it is invisible from inside the suite, because the
suite's author and the suite share the same blind spots.
This skill attacks that gap from two directions at once, then makes both
directions earn their findings:
| Pass | What it sees | What it asks |
|---|
| Adversarial | source + tests | "Where does a plausible mutant survive green?" |
| Cold read | source only — tests hidden | "What coverage would you demand here?" |
| Verify | the actual code | "Is each finding real, or did the model invent it?" |
The cold read is the move with no other home, and the reason this skill
exists. Everything else composes skills you already have.
When to Use This Skill
| Use this skill when... | Use something else instead when... |
|---|
| A suite is green but you doubt it would catch a real regression | You need to write tests → testing-plugin:test-setup / language test skills |
| Stakes are high: ML numerics, serialization, concurrency, config→behavior wiring | You want a mutation tool run → testing-plugin:mutation-testing |
| Before trusting a suite as the gate for a risky migration | You want test-smell/flake heuristics → testing-plugin:test-quality-analysis |
| A suite grew organically and nobody has audited what it doesn't cover | Adversarially reviewing code/a design (not a suite) → agent-patterns-plugin:adversarial-review |
| The suite is small, low-stakes, or throwaway — this manufactures busywork |
The Three Passes
1. Adversarial — hunt the surviving mutant
Reuse the posture from agent-patterns-plugin:adversarial-review (inverted
objective + triage gate); this skill only supplies the test-suite lens. Give
the reviewer source tests, and make it hunt concretely:
