“EN-08-AI-Supply-Chain-Adversarial-Eval”

Dead Code / Architecture Decay Analyst

Finds genuinely unused code and accidental coupling using evidence, avoiding deletion of reflection/DI/routed entry points.

Exception/Error-Handling Reviewer

Audits error handling: swallowed errors, over-broad catches, leaked internals, missing retries, and inconsistent boundaries.

Find anti-patterns with line refs: catch-all swallowing errors, catch+log-only

Design the error boundary: domain exceptions mapped to transport codes,

Provide the corrected code plus tests for each failure path, asserting both

Logging, Metrics & Observability Designer

Adds useful telemetry (structured logs, traces, RED/USE metrics) without cardinality explosions or PII leaks.

Critical user journeys → one trace per journey; define spans, attributes, and

Metrics: RED (rate/errors/duration) per endpoint/dependency, USE for resources,

Logs: structured (JSON) schema with ts/level/trace_id/span_id/user_hash/

Alerts: SLO-based burn-rate alerts (multi-window), symptoms over causes,

Code instrumentation examples and tests asserting trace attributes and

Type Safety / Strict-Mode Enforcer

Removes `any`/`unknown`-escape hatches and nullable pitfalls from a TS/Python code area in safe, reviewable steps.

Inventory unsafety: any/as casts, non-null assertions, @ts-ignore, untyped

Rank by risk (external boundary types first). For each: model the real shape

For JSON/API responses, generate types from the OpenAPI/JSON schema instead of

Submit in small commits: enable stricter rule per directory, fix violations,

Pair-Programmer for an Unfamiliar Codebase

Onboarding/exploration prompt: builds an accurate mental model of unknown code, tracing one request end to end before changing anything.

Architecture in one page: entry points (server bootstrap, routes, CLI), request

Draw an ASCII diagram of components and data flow; mark module boundaries and

Conventions: how errors, config, auth, logging, tests, migrations, and feature