Engineering review using AI coding tools
This review was performed in the same implementation session using AI coding tools. It is not independent human certification or an n8n runtime audit.
Reviewed JSON graph connections and settings, embedded Code-node sources, the policy linter, mutation tests, fake receiver contract, recorded replay outcomes and handoff instructions. Focus: finite graph/input, string event identity, canonical payload conflict checks, bounded HTTP policy, original-item linking for dead letters, credential references, failure visibility and recovery evidence.
The Code-node sources are tested directly from the exports, and their equality with reviewed source files is asserted. A development comparison failure was traced to the host/VM array boundary in the test harness; results are now normalized as serialized data before comparison. No runtime defect was masked by weakening the event-identity assertion. Static linter findings never print the sensitive header value, including the known fake baseline marker.
Residual limits: policy checks are not a complete n8n schema or secret scanner; graph/source patterns cannot prove arbitrary code behavior; real scheduling, item linking, error-trigger delivery, HTTP retry timing, activation and durable receiver/storage contracts need staging acceptance. The replay's receiver and dead-letter store are in memory. Header idempotency only works with an atomic receiving service. See evidence/tests.json, evidence/lint.json, evidence/replay.json and HANDOFF.md.