sdlc-verify

# sdlc-verify **Executor resolution (host-aware):** - **Claude Code + `superpowers` installed** → prefer **`superpowers:verification-before-completion`** (its richer version). - **Otherwise** (Cursor / any host / no companion) → use this. Same discipline, portable. ## The rule **Evidence before claims — always.** Claiming work is complete without verifying it isn't efficiency, it's dishonesty. **No completion claim without fresh verification output in this turn** — if you didn't run it just now, you can't claim it. (Different words don't exempt you: spirit over letter.) ## The gate — run this before ANY "done / fixed / passing" 1. **Identify** the command that proves the claim — the repo's tests/build/lint, or `.sdlc/project.md` → **Verify command**. 2. **Run** it fresh and in full. 3. **Read** the actual output — exit code, failure count. 4. **Verify** the output confirms the claim. If it doesn't, state the *real* status *with* the output — don't hedge. 5. **Only then** make the claim, and make it *with* the evidence. ## What each claim actually requires | Claim | Proof required | Not enough | |---|---|---| | Tests pass | test output: 0 failures | "should pass" / a previous run | | Lint / type clean | linter output: 0 errors | a partial check | | Build works | build: exit 0 | "the linter passed" | | **Bug fixed** | reproduce it (**red**) → fix → it's gone (**green**) | "changed the code, assume fixed" | | A subagent finished | check the diff yourself | the agent's ...