inspect-bdd-steps

# BDD Step Assertion Completeness Inspector Inspects every BDD step function for semantic completeness: does the function body actually implement what the step text claims? ## Usage ``` /inspect-bdd-steps /inspect-bdd-steps --pass1-only ``` ## What It Does 1. **AST scan**: Extracts all `@given`/`@when`/`@then` decorated functions from `tests/bdd/steps/` 2. **Pass 1 (Sonnet triage)**: For each Then step, asks: "Is there a HIGH chance this function does NOT implement what the step text claims?" - PASS: Function plausibly implements its claim - FLAG: Function likely doesn't (pass body, truthiness-only check, etc.) 3. **Pass 2 (Opus deep trace)**: For each FLAG, collects production context (schemas, error classes, harness code) and asks Opus to make an architectural judgment about what the correct assertion should be. 4. **Report**: Writes `.claude/reports/bdd-step-audit-<date>.md` with findings grouped by severity (MISSING > WEAK > COSMETIC). ## Protocol Run the inspection script: ```bash python3 .claude/scripts/inspect_bdd_steps.py ``` Options: - `--pass1-only` — Skip Pass 2 deep trace (fast triage only) - `--steps-dir PATH` — Override step definitions directory - `--output PATH` — Override report output path - `--then-only` — Only inspect Then steps (default: true) Review the generated report and use findings to inform beads-58u (assertion mismatch fixes) or create new tasks. ## When to Use - After writing new BDD step definitions - After modifying existing step assertions - As a periodic audit (monthly or per-epic) - Before closing BDD-related PRs