assertion-quality

# Assertion Diversity Analysis Agent evaluates whether tests prove behavior with meaningful and varied assertions. ## When to Use | User prompt | Use | |---|---| | User asks whether assertions are too shallow | Use this skill | | User asks for zero-assert or trivial-assert detection | Use this skill | | User asks for suite-wide assertion category balance | Use this skill | ## When Not to Use | User prompt | Route | |---|---| | User asks to write or repair tests | Use a test-writing skill | | User asks for general suite defects beyond assertions | Use `test-anti-patterns` | | User asks for runtime coverage | Use `coverage-analysis` or `test-gap-analysis` | ## Required Inputs | Input | Required | Description | |---|---|---| | Test scope | Yes | Agent needs files, classes, projects, or named tests. | | Production context | No | Agent reads it when business significance is unclear. | | Framework | No | Agent infers it, then reads `test-analysis-extensions`. | ## Assertion Category Table | Category | Agent counts | Weak signal | |---|---|---| | Equality and comparison | Expected values, ranges, ordering, tolerances | Many equality checks only | | Boolean and null | Presence, absence, flags, state guards | Tautologies or null-only suites | | Exception and negative | Expected failures or forbidden paths | Broad or detail-free exception checks | | Type and structure | Runtime type, object shape, snapshots, collection contents | Stringified structure only | | State and side effect | Mutations, persistence, messages, collaborator interactions | Verification of setup rather than outcomes | ## Quality Pattern Table | Pattern | Signal | Agent fix | |---|---|---| | Zero-assert test | No assert, verify, or snapshot exists | Agent flags false confidence. | | Trivial-only test | Assertions prove existence only | Agent recommends behavior-specific checks. | | Single-style suite | Nearly all tests use one assertion category | Agent recommends missing categories. | | Happy-path-only verification | Negative or exceptional outcomes are absent | Agent recommends failure-path assertions. | | Structural-blind test | Complex outputs receive scalar checks only | Agent recommends structural or state assertions. | ## Workflow | Step | Agent action | Test | Pass | |---|---|---|---| | 1 | Agent reads framework guidance. | Review extension file. | Framework-specific assertion forms are recognized. | | 2 | Agent inventories assertion forms in scope. | Review extracted assertions. | Every test has a counted verification style or a zero-assert note. | | 3 | Agent groups assertions by category and depth. | Review category map. | Each assertion lands in one primary category. | | 4 | Agent calculates concise metrics. | Review metrics table. | Output includes zero-assert, trivial-only, and category spread. | | 5 | Agent reports gaps without confusing volume for strength. | Review findings. | Duplicate equality checks do not inflate diversity claims. | ## Output Contract | Output | Requirement | |---|---| | Suite summary | Agent reports counts and category spread. | | Weak tests | Agent identifies zero-assert and trivial-only cases. | | Missing categories | Agent names underused negative, structural, or state assertions. | | Recommendations | Agent gives category-targeted next steps. | ## Verification Checklist - [ ] Agent loaded framework guidance before classifying assertions. - [ ] Agent counted mock verifications and snapshots when they prove behavior. - [ ] Agent separated trivial checks from meaningful single-assert tests. - [ ] Agent reported diversity by category, not raw volume only. - [ ] Agent connected recommendations to specific tests or suite gaps. ## Common Pitfalls | Pitfall | Agent fix | |---|---| | Agent treats every boolean assertion as trivial | Agent flags only tautological or low-signal boolean checks. | | Agent penalizes valid exception-only tests | Agent treats strong exception assertions as complete for that intent. | | Agent misses framework-specific assertion helpers | Agent reads the extension file first. | | Agent calls high-volume equality suites diverse | Agent measures spread across categories. |