← ClaudeAtlas

verifylisted

Ruthless verification — closed feedback loop until green, regression tests, adversarial review in a fresh context, gap-closure audit, and an evidence block. Use after any implementation, before declaring any task done, or when the user asks "is this actually done/correct/tested".
omeeragtoprak/agentic-engineering-protocol · ★ 0 · Code & Development · score 72
Install: claude install-skill omeeragtoprak/agentic-engineering-protocol
# AEP Phase 4 — Verify **The prime directive: YOU MUST NOT declare a task complete while any check is failing — and completion claims without evidence are void.** ## 1. The verification loop Run, in order: **build/compile → linter + type-checker → affected tests → full suite when feasible.** Then iterate: ``` edit → run check → READ the output → fix root cause → re-run ``` - Every warning, type complaint, and failure is a blocker: fix it or explicitly justify it in the evidence block. - **Two-strike rule:** two consecutive iterations without measurable progress = STOP. Reassess the hypothesis, return to `aep:plan`, or escalate. Never suppress an error to make the loop pass (no skipped tests, no lint-disable comments, no `catch {}`) — that is falsifying evidence. ## 2. Tests for new logic - New logic requires new tests: happy path + boundaries + failure modes. - Bug fixes require a **regression test that fails before the fix and passes after** — run it both ways and show it. - Tests must be deterministic and order-agnostic; mock only at architectural boundaries; never mock the unit under test. - **Test integrity:** never weaken, skip, or delete an existing test to get the gate green — the gate reports uncommitted test-file changes, and any legitimate test change (rename, strengthened assertion, new case) must be named and justified in the delivery summary. ## 3. Adversarial review (fresh context) The agent that wrote the code does not grade it. For significant diffs: