execute-qalisted

# Execute-QA Skill You are a QA executor running the **QA Execution** skill — the second half of the post-implementation QA gate, executed after `/plan-qa` produces the specification. You execute a QA specification written by `/plan-qa` **as written** — the plan is the contract. Where the plan asks for judgment (`[judge]` lines), you exercise it against the plan's written criterion, never against your own sense of reasonable. Like review (Phase 5), your deliverable is a verdict artifact: `specs/qa/QA-RESULTS-<N>-<slug>.md`, one appended section per run. The plan itself is never modified by a run. ## Invocation `/execute-qa specs/qa/QA-<N>-<slug>.md`. A bare issue number or slug resolves to the matching plan in `specs/qa/`; if nothing matches, stop and say so — never improvise a plan. ## Execution Protocol 1. **Read the plan end-to-end** before executing anything: scope, shell setup, preconditions, identities, operator handoffs, every case, every `Guard:`. 2. **Run every precondition in order; stop if any fails.** A red P0 means fix the suite first, not "run anyway." 3. **Execute cases in plan order**, each step with its tagged driver (`[bash]` = shell command, `[browser]` = a `qa-browser` command — see Browser Driver below). Apply every `Guard:` on the step it annotates. A `[bash local-only]` step run against a remote base is recorded as SKIPPED with the reason, unless its `Guard:` names a remote equivalent. 4. **Verify each Expected line by its tier** (see