Testing & Debugging for Claude Code

Retroactive audit of an implemented AI phase's evaluation coverage. Checks implementation against the AI-SPEC.md evaluation plan. Scores each eval dimension as COVERED/PARTIAL/MISSING. Produces a scored EVAL-REVIEW.md with findings, gaps, and remediation guidance. Spawned by /gsd:eval-review orchestrator.

Verifies cross-phase integration and E2E flows. Checks that phases connect properly and user workflows complete end-to-end.

Fills Nyquist validation gaps by generating tests and verifying coverage for phase requirements

Verifies phase goal achievement through goal-backward analysis. Checks codebase delivers what phase promised, not just that tasks completed. Creates VERIFICATION.md report.

Tier 1 (surface) Ralph Review. Dispatched by the /ralph-review command. Runs fast, cheap surface validation against a finished implementation diff. Never writes code.

Multi-service infrastructure health checker. Probes every AWS service the caller has IAM access to (ECS, EC2, RDS, Lambda, S3, CloudFront, ALB/NLB, API Gateway, SQS, SNS, DynamoDB, ElastiCache, Route 53, ACM, CloudWatch, Budgets, IAM) plus Vercel and GitHub Actions. Returns structured JSON with service health, accessible/inaccessible service lists, and severity-ranked anomaly flags. Used by ops-fires and ops-deploy.

Lightweight APM and metrics probe agent. Queries Datadog, New Relic, and OpenTelemetry backends for active alerts, error traces, and entity health. Returns structured JSON. Read-only — used by ops-monitor skill.

Use this agent to autonomously build, test, and deploy complete applications from plain-English descriptions. Runs a 9-phase pipeline across 4 stacks with enterprise-grade safety hooks.

Tier 2 (logic) Ralph Review. Dispatched by the /ralph-review command after Tier 1 passes. Runs medium-depth logic validation — evidence quality, scope alignment, fail-fast, observability, cross-boundary contracts. Never writes code.

Investigates a specific issue from Sentry, Linear, or GitHub. Finds the root cause in code, checks if it's already fixed, and either confirms resolution or creates a fix branch with a PR.

Agent that verifies build, lint, test, and checklist after TASK completion within a WORK. Automatically invoked by the scheduler. Verifies in read-only mode without modifying code.

Detects traffic spikes, drops, bot activity, and tracking gaps in Umami analytics data, then classifies each anomaly by severity and root-cause hypothesis. Use when investigating unexpected traffic changes or data quality issues. Trigger with \"check for anomalies\", \"why did traffic drop\".

Detect data quality issues, stubs, orphan plugins, duplicate files, and score outliers in the freshie inventory SQLite database using a structured six-check sweep. Use when auditing inventory health after a discovery run or investigating anomalous compliance grades. Trigger with \"scan for anomalies\", \"detect inventory issues\".

Diagnoses REST API failures by comparing HTTP logs against OpenAPI specs, identifies root cause by status code category, and generates working cURL repro commands. Use when an API call returns unexpected errors or status codes. Trigger with "why is my API failing", "debug this API error".

Use when reviewing HTTP/RPC/GraphQL endpoints, request/response contracts, input validation, or error semantics — verifies the API surface against the api persona standards.

Generates and executes REST and GraphQL API test suites covering happy paths, auth scenarios, edge cases, and contract validation, then reports results with coverage metrics. Use when automating endpoint regression testing or building a new test suite. Trigger with "test this API", "generate API tests".

Scans a codebase for dangerous async patterns — floating promises, async forEach, missing await, unhandled rejections, and mixed styles — with confidence scoring and remediation guidance. Never auto-applies fixes. Use when auditing Node.js or TypeScript code for async safety issues. Trigger with "find async issues", "scan for floating promises".

Use this agent for bd/Dolt incident response — a dolt-server that won't start or has orphaned, server sprawl, suspected lost writes after rapid bd updates, JSONL that lags the database, or migrating a workspace between embedded and server mode. It knows the rapid-write race is already fixed in bd 1.0.4 and that residual lag is only the JSONL export throttle.

Design API benchmarks, profile p50/p95/p99 latency, set up throughput tests, and detect performance regressions with k6/wrk. Use when establishing baselines or catching latency regressions in CI. Trigger with "benchmark this API", "set up a performance test".

Parse, classify, redact PII, score reliability, and cluster bug candidates by family and signal layers. Use when processing raw X/Twitter posts into structured bug clusters.

Designs controlled failure experiments to expose resilience gaps before production incidents do. Use when you need a chaos experiment hypothesis, a game day plan, or a blast-radius-limited resilience audit. Trigger with \"design a chaos experiment\", \"plan a game day\".

Designs and executes controlled chaos experiments — failure injection, latency simulation, resource exhaustion — following the scientific method with defined abort conditions and recovery validation. Use when hardening a system against cascading failures or preparing for GameDays. Trigger with "design chaos experiment", "test system resilience".

Detects circular module dependencies using madge/dependency-cruiser, classifies each cycle as runtime-critical or type-only, and proposes minimal-blast-radius resolution strategies. Never auto-applies fixes. Use when debugging mysterious initialization failures or reducing bundle size. Trigger with "find circular dependencies", "untangle module cycles".

Designs data validation, cleaning, and quality-monitoring pipelines so models train on trustworthy data. Use when you need deduplication logic, outlier detection, or an ETL quality gate. Trigger with \"audit my data quality\", \"design a cleaning pipeline\".

Expert UX evaluator for CLIs and developer APIs. Synthesizes pre-collected test data into an 11-criteria evaluation and writes artifacts to a timestamped directory. Launched by the cli-ux-tester skill.

Triage a failed or slow Databricks cluster by fanning out parallel root-cause threads — one per cause class (provisioning, network/NPIP, init-script, driver, spot/shuffle) — over the cluster's live event stream, then aggregate to the single most-likely cause with its evidence. Use when a cluster failed to start, died mid-run, or is inexplicably slow. Trigger with "why did this cluster fail", "investigate cluster", "triage cluster start".

Run enterprise-tier skill compliance validation, populate the freshie SQLite database, and report grade distribution, average score, worst offenders, and B-grade upgrade candidates. Use when measuring overall marketplace quality or identifying skills that need remediation. Trigger with \"run compliance validation\", \"grade the inventory\".

Generates realistic, locale-aware test data (users, products, orders, custom schemas) using Faker.js, Factory Boy, or json-schema-faker — producing factory functions, database seed scripts, and fixture files ready for immediate use. Use when setting up a test environment or populating a dev database with production-scale data. Trigger with \"generate test data\", \"create seed data factories\".

Use when database performance is in question — query optimization, indexing strategy, query plans, schema/access-pattern tradeoffs, or "is this the fastest way to read/write this data". The DB performance specialist (distinct from database-reviewer, which owns migration correctness/reversibility). Verifies against the db and performance persona standards.

Use when reviewing schema changes, migrations, indexes, or query patterns — verifies reversibility, index strategy, and query-plan correctness against the db persona standards.

Scans for unused exports, dead imports, unreachable code, and stale feature flags using knip/vulture/deadcode, auto-removes high-confidence findings after build verification, and flags the rest for manual review. Use when cleaning up a codebase before a refactor or release. Trigger with "find dead code", "remove unused exports".

Use when finding the root cause of a bug, failing test, or regression — runs systematic 5-Whys and parallel hypothesis testing before any patch. The root-cause investigator behind /hyperflow:trace.

Scans TypeScript/JavaScript for unnecessary null checks, impossible try/catch blocks, redundant validation, and dead catch blocks — tracing data flow to prove each defense is unneeded before flagging. Use when you want to clean up defensive-programming noise after strictNullChecks is on. Trigger with \"find unnecessary null checks\", \"audit defensive code\".

Use when reviewing CI/CD pipelines, infrastructure-as-code, container/build config, deployment, or observability changes — verifies operational safety against the devops persona standards.

Designs statistically rigorous experiments — A/B tests, power analysis, variance reduction, and causal inference frameworks. Use when designing an experiment, analyzing test results for validity, or auditing experimentation infrastructure. Trigger with \"design A/B test\", \"review experiment methodology\".

Builds LLM eval harnesses — benchmark suites, automated regression pipelines, golden set management, and human eval orchestration. Use when measuring model quality, wiring evals into CI, or auditing benchmark validity. Trigger with \"design eval harness\", \"build LLM regression suite\".

Triages and resolves FairDB PostgreSQL production incidents using a P0–P3 severity protocol — diagnosing disk, connection, and performance failures, then generating incident reports and stakeholder communications. Use when a FairDB database is down or degraded. Trigger with "database is down", "FairDB incident".

Generates and validates production-grade Firestore security rules for user auth, role-based access, A2A service accounts, and data validation patterns, with emulator-ready unit tests. Use when securing Firestore collections or enabling agent-to-agent communication. Trigger with "generate Firestore security rules", "secure my collections".