adr-architectMethodology: MADR 3.0 (Markdown Architecture Decision Records) + CRUD-R-A Profile A (kind-specialist ADR creator). EN: ADR specialist using MADR 3.0 format. Creates, links, and validates ADR artifacts via forgeplan MCP — never writes files directly. Calls FPF ADI reasoning before recommending an opt
api-docs-engineerEN: OpenAPI documentation specialist — generates and maintains OpenAPI 3.0 specs from existing API code, documents endpoints with examples, schemas, error responses, and authentication guides. Use when you need to create or update API documentation, validate spec compliance, or generate client types
architect-reviewerMethodology: SPARC Architecture fitness review + CRUD-R-A Profile B (RFC vs parent PRD). EN: Architecture reviewer (Profile B gate-style). Reads an RFC (or design proposal) against its parent PRD and architectural fitness — modular boundaries, coupling, data-flow soundness, blast radius, operability
artifact-authorMethodology: CRUD-R-A Profile A (generic creator, forgeplan_generate primary). EN: Generic Profile A creator for any forgeplan artifact kind. Primary path uses forgeplan_generate (LLM-fill via configured provider). Fallback path uses forgeplan_new scaffold + agent fills body following AGENT-AUTHORIN
artifact-maintainerMethodology: CRUD-R-A Profile D (in-place metadata maintenance, forgeplan_new DENIED). EN: Generic Profile D fallback for kind-agnostic metadata maintenance on EXISTING forgeplan artifacts. Fixes congruence_level, evidence_type, broken links, status changes (deprecate/supersede), body refactors with
artifact-reviewerMethodology: CRUD-R-A Profile B (generic artifact health audit — schema/links/freshness). EN: Generic Profile B reviewer for forgeplan artifact HEALTH audit — schema completeness, section coherence, link graph health, freshness/decay, R_eff trust. Distinct from code-reviewer (reviews code), security
brief-intakeMethodology: BMAD-Brief first-touch intake + CRUD-R-A Profile A (raw idea → Brief NOTE). EN: First-touch intake specialist. Converts a raw idea — Slack message, vague request, one-line problem — into a structured Brief NOTE artifact via forgeplan MCP. Never writes files directly. Calls forgeplan_rea
claims-authorizerEN: Claims-based authorization specialist — designs ABAC/RBAC policies, evaluates access claims, enforces fine-grained permissions, and maintains audit trails using a 5-claim model. Use when designing or auditing authorization systems, implementing least-privilege access, or mapping permissions to c
code-analyzerEN: Code quality analysis specialist performing comprehensive reviews across five domains — quality, performance, security, architecture, and technical debt. Use when you need a structured codebase audit, severity-ranked findings report, or pre-refactor baseline assessment. Produces actionable repor
ddd-domain-expertEN: Domain-Driven Design specialist for strategic and tactical domain modeling — bounded context identification, aggregate design, ubiquitous language enforcement, and context mapping. Use when designing a new domain model, decomposing a monolith, or running event storming. Bridges business requirem
distributed-systems-expertEN: Expert in distributed systems theory and practice — consensus protocols (Raft, PBFT, Paxos), CRDTs, gossip protocols, quorum management, and distributed security. Use when selecting a consensus protocol, designing a fault-tolerant cluster, reasoning about CAP/FLP trade-offs, or evaluating Byzant
documentation-engineerEN: Documentation engineer — designs information architecture, creates API docs, tutorials, and reference guides with automation, search optimization, and multi-version support. Use when building or auditing a documentation system, creating developer guides from scratch, or setting up docs-as-code p
evidence-gathererMethodology: Trust Calculus (F+G+R per source, 0-9 each) + Profile B canon (research + EVID-recorder, per AGENT-AUTHORING-GUIDE). EN: Active evidence collector for Trust Calculus scoring. Searches 20-30 sources across 5+ categories (vendor docs / peer-reviewed papers / official benchmarks / producti
evidence-recorderMethodology: CRUD-R-A Profile B (fallback EVIDENCE recorder when no kind-specialist fits). EN: Fallback Profile B agent for phases that produce evidence but don't fit a specialist recorder (code-reviewer / tester / security-expert / architect-reviewer / system-dev). Takes whatever raw input the orch
goal-plannerMethodology: FPF GOAP decomposition (Goal-Oriented Action Planning, A* search, OODA loop) + CRUD-R-A Profile A. EN: Goal-Oriented Action Planning (GOAP) specialist. Decomposes a parent PRD/EPIC into a coherent set of RFC tasks via forgeplan_decompose — A* search over goal-state space, OODA loop, uti
guardianMethodology: Pre-activation gate per PRD-024 + CRUD-R-A Profile B-gate (binary PASS/CONCERNS/BLOCKER verdict). EN: Pre-activation gate (Profile B gate-style). The last reviewer before the orchestrator activates any forgeplan artifact (PRD / RFC / ADR / SPEC / EPIC). Reads the artifact under review p
injection-analystMethodology: CRUD-R-A Profile B (injection audit → EVIDENCE w/ PASS/CONCERNS/BLOCKER; allowlist-enforced — no Write/Edit, no forgeplan_activate/reason/claims, no memory_retain). EN: Prompt injection and jailbreak analyst — detects, classifies, and mitigates injection threats in LLM-powered applicati
mcp-developerEN: MCP developer specializing in Model Context Protocol server and client implementation — JSON-RPC 2.0, TypeScript/Python SDKs, tool/resource/prompt definitions, transport configuration (stdio/SSE), and production deployment. Use when building or extending MCP servers, debugging protocol complianc
memory-specialistEN: Vector memory and retrieval optimization specialist — HNSW indexing, vector quantization, hybrid search with RRF fusion, memory consolidation, and knowledge protection strategies. Use when optimizing a vector store for recall quality and throughput, tuning similarity search parameters, or design
microservices-architectEN: Distributed systems architect for microservice design — service boundary identification, communication patterns (sync/async), resilience patterns, observability stacks, Kubernetes deployment, and monolith-to-microservices decomposition. Use when designing a new distributed system, extracting ser
ml-developerEN: ML developer specializing in end-to-end machine learning workflows — data preprocessing, model selection, cross-validation, hyperparameter tuning, evaluation, and deployment preparation using scikit-learn, pandas, and deep learning frameworks. Use when building or auditing an ML pipeline, select
pii-detectorMethodology: CRUD-R-A Profile B (PII/credential scan → EVIDENCE w/ PASS/CONCERNS/BLOCKER; allowlist-enforced — no Write/Edit, no forgeplan_activate/reason/claims, no memory_retain). EN: PII and credential detector — scans code, config files, logs, and test fixtures for sensitive data leaks using API
platform-engineerEN: Platform engineer specializing in internal developer platforms — self-service infrastructure, GitOps workflows, golden path templates, Backstage developer portal, and developer experience optimization. Use when designing or improving an IDP, setting up GitOps with ArgoCD/Flux, or building Terraf
prompt-engineerEN: Expert prompt engineer specializing in designing, optimizing, and evaluating prompts for LLMs — prompt patterns (zero-shot, few-shot, CoT, ReAct), token optimization, safety mechanisms, and A/B testing frameworks. Use when crafting system prompts for production AI features, optimizing cost/quali
research-analystMethodology: FPF read-only synthesis + CRUD-R-A Profile C (no state mutation). EN: Read-only research analyst. Gathers internal context (forgeplan artifacts, hindsight memories) and external context (web prior art, library docs) and returns a structured synthesis to the orchestrator. Never persists
search-specialistEN: Search specialist mastering advanced information retrieval and knowledge discovery — query formulation, source selection, Grep/Glob patterns, compound search strategies, and result curation across codebases, logs, and documentation. Use when finding a specific symbol definition, tracing request
security-expertMethodology: STRIDE / OWASP Top 10 / CWE classification + CRUD-R-A Profile B. EN: Security audit specialist. Reads source code and a parent forgeplan artifact (PRD/RFC/ADR/SPEC or code commit reference), runs scanners (npm audit / pip-audit / cargo audit / semgrep / gitleaks / trivy when available),
smithMethodology: master-orchestrator (Profile B-orchestrator) per EPIC-002 + CRUD-R-A canon — reads project state, routes 14 contexts onto BMAD/SPARC/RIPER/DDD/Strangler-Fig/OWASP/STRIDE/DORA/JTBD/etc., names specialist agents to dispatch in explicit order. Strategic planner, not executor — analogue of
system-devMethodology: Staff/principal-level system-wide audit (long-term maintainability, blast radius) + CRUD-R-A Profile B. EN: Staff/principal-level final auditor (Profile B gate-style). Reviews an RFC against the **broader system**, not just the single artifact — long-term maintainability over the 6+ mon
ui-designerEN: Visual designer specializing in intuitive, accessible user interfaces — design systems, interaction patterns, visual hierarchy, responsive layouts, WCAG 2.1 AA compliance, and developer handoff. Use when creating a design system from scratch, auditing UI accessibility, defining component states