verification-before-completionlisted

# Verification before completion ``` NO COMPLETION CLAIMS WITHOUT FRESH VERIFICATION EVIDENCE ``` If you haven't run the check in this session after your last change, you can't claim it passes. ## The gate 1. **Identify**: which command proves the claim? Use the project's commands (the kit verify script, CLAUDE.md, `.claude/verify.json`, CI config). 2. **Run** the full command, fresh. Don't use a filtered subset, and don't pipe it into `tail` in a way that hides the exit code. 3. **Read** the output: exit code, count of failures, warnings. 4. **Verify**: does the output confirm the claim? If not, report the actual state with the evidence. 5. **Only then claim it**, and quote the evidence ("`npm test`: 142 passed, 0 failed"). | Claim | Requires | Not enough | |---|---|---| | Tests pass | Full test run, 0 failures | An earlier run, "should pass", one file | | Builds | Build command exits 0 | Lint or typecheck passing | | Bug fixed | The original symptom no longer reproduces, plus a regression test | "Code changed" | | Requirement met | Checked against each numbered criterion | "Tests are green" | | Subagent finished | You inspected the diff and ran the checks | Its "DONE" report | | Works on the device or UI | You exercised it (emulator, browser, screenshot) | Unit tests alone | ## Report honestly The final message lists: - **Changed:** files, one line each. - **Checks run:** command → result, only those actually run in this session. - **Criteria:** each requirement or c