An engineering process for Claude Code projects: plan, review, record. The commit gate is a git hook, so it isn't optional.
Agent Claude Code
Use proactively to run this repo's test suites and g-dev/ gate fixtures and return attested runner output. The execution counterpart to test-writer, which has no execution tool and cannot run anything — g-forge-dev closes that gap for THIS repo's own dogfooded plugin code only.
Agent
Use proactively when files in layer-boundary directories change. Validates import direction, circular deps, and separation of concerns. Reports violations with file:line refs. Does not fix.
Agent
Use before any merge and when project-manager needs technical risk assessment. Guards milestone feasibility, checks done conditions, and reviews diffs directly. Does not implement.
Agent
Use proactively after any code change and before every merge. Reviews for logic errors, code smells, DRY violations, and edge cases. Reports with file:line refs and severity. Does not fix.
Agent
Use proactively when a bug resists a first fix attempt. Reproduces the failure, traces root cause, and proposes a fix strategy. Does not implement.
Agent
Use proactively before any release and whenever the dependency manifest changes. Audits for security advisories, deprecated packages, license conflicts, and unused declarations. Does not fix or upgrade.
Agent
Use proactively when documentation changes, when public exports change, or at milestone close. Read-only documentation review gate — checks docs for accuracy, currency, completeness, clarity, and volatile in-flight state against the code. Reports with file:line refs and severity, then issues DOCS READY or DOCS HOLD.…
Agent
Use proactively after implementation is complete or when exports lack documentation. Writes inline docs and README sections explaining WHY, not WHAT.
Agent
Use proactively when facing cryptic errors or production incidents before attempting a fix. Parses logs and stack traces to identify patterns and root causes. Does not fix.
Agent
The generic, stack-agnostic wave implementer — the default executor for any implementation task that has no matching stack implementer installed by /g-specialize, and the fallback for projects that have not been specialized. Implements one wave task to its done condition. Single-use — one approach, one attempt.…
Agent
Use proactively on performance-sensitive changes. Flags O(n²) paths, N+1 queries, unnecessary re-renders, and hot-path waste. Reports with file:line refs and estimated impact. Does not fix.
Agent
Use proactively before opening a pull request. Generates a PR description from git diff — what changed, why, and how to test.
Agent
The user-facing interface for every session. The user talks to the PM — not to a neutral assistant. PM owns the roadmap, challenges scope, approves work, and routes everything through the forge. Does not write code or touch implementation files.
Agent
Use when a refactor spec from spec-writer is ready to execute. Executes exactly as written — no scope creep, no judgment calls.
Agent
Coordinates the full review pipeline — code review, architecture, security, and performance in parallel. Aggregates findings into one report. Does not review itself. Must run as the root session agent (--agent review-orchestrator or directly from a skill in the main session) — spawning it as a nested subagent prevents…
Agent
Use proactively before any merge, especially those touching auth or external integrations. Audits for OWASP Top 10 vulnerabilities, injection vectors, and secrets exposure. Does not fix.
Agent
Use when a task needs to be specced before handoff to an executor. Produces an implementation spec precise enough for a Haiku agent to execute without judgment calls.
Agent
Use at the start of any multi-step implementation before touching code. Breaks the request into atomic, verifiable tasks with done conditions.
Agent
Use proactively after implementing code that needs coverage or when spec-writer produces a spec. Writes unit, integration, or e2e tests. Fixed data only, no Date.now() or random values.
Agent
Use immediately after task-decomposer. Takes a task list and produces a parallel wave execution schedule by mapping dependencies, and tags each task with the executor agent that should run it.
Agent
Angular 17+ standalone components architecture specialist. Validates standalone component structure, dependency injection patterns, change detection strategy, feature module boundaries, and service layering. Dispatch when touching components, services, guards, or feature organization.
Agent
ASP.NET Core 8 architecture specialist. Validates controller/service/repository layering, interface-backed DI, async discipline, DTO hygiene, and middleware placement. Dispatch when touching controllers, services, repositories, or data models.
Agent
Astro architecture specialist. Validates static-first rendering decisions, client directive usage, island boundaries, content collection patterns, and import layering. Dispatch when touching pages, components, islands, or content.
Agent
Bun + Elysia architecture specialist. Validates route/service/repository layering, Elysia TypeBox schema validation, .derive() guard patterns, ESM-only discipline, and typed context. Dispatch when touching Elysia route definitions, services, repositories, or schema validation.