Skip to content

Agent Catalog Overview

FlowForge ships with 36 active specialist agents, each purpose-built for a narrow domain. The Maestro orchestrator never writes code directly. Instead, it consults these specialists, presents options to the developer, and coordinates the specialists chosen to execute the work.

This page introduces the catalog by category. Every agent has a dedicated detail page with its capabilities, typical use cases, and example prompts. See the full table on the catalog index.

All 36 agents are available today as part of the FlowForge core distribution. Some capabilities may move to Pro or Team tiers in future releases (team-wide analytics, managed deployments, advanced auditing), but the agents themselves — the specialist prompts that power FlowForge’s orchestration — are and will remain free to use.

The phase-0 specialist dispatched first on any legacy, existing, or unfamiliar codebase. It reads cheaply on the Haiku model tier so expensive specialist tokens are spent only on verified knowledge.

  • fft-code-explorer — Haiku-class ingest: source tree scan, lockfile version extraction, schema inference from migrations (no live creds), framework detection, and structured specialist-wave dispatch. Produces a laudo + ERD + FF-branded diagrams + a filled project CLAUDE.md. ForgePlay #1 entry point.

These agents shape the what and the why before any code is written.

  • fft-requirements-analyst — anti-fabrication requirements interrogation, outside-in interrogation, 8-artifact handoff packet. Dispatch before fft-architecture on greenfield work.
  • fft-architecture — system design, ADRs, SOLID principles, microservices and monolith trade-offs.
  • fft-project-manager — task breakdown, sprint planning, milestone tracking, GitHub structure.
  • fft-product-owner — requirements, user stories, backlog prioritization, stakeholder alignment.

Server-side craft, from contract design to persistence.

  • fft-api-designer — REST, GraphQL, gRPC, WebSocket; OpenAPI specifications; versioning and contract testing.
  • fft-backend — Node.js, Python, Ruby, Java, Go; scalable server architectures with 80%+ test coverage.
  • fft-database — schema design, migrations, query optimization, ACID and data-strategy decisions.

Everything the user sees and touches — led by the design authority that briefs every surface.

  • fft-design-architect — master design authority: holds the guideline corpus (Apple HIG, Material 3, WCAG 2.x, FlowForge design canon) and emits the per-work design brief every leaf implements. Dispatch first on design-strategy and multi-surface work.
  • fft-frontend — React, Vue, Angular; component architecture, accessibility, performance.
  • fft-designer — web + print/editorial (papelaria) UI/UX: the Manual, PDF-from-HTML, reports, and stationery as well as web surfaces; design systems, type-scale systems, page grids, print stylesheets.
  • fft-brand-architect — brand identity, color theory, typography, design tokens, guidelines.

Platform-native and cross-platform — three specialists, one per platform.

  • fft-ios — Swift, SwiftUI, UIKit, structured concurrency.
  • fft-android — Kotlin, Jetpack Compose, Hilt, Room.
  • fft-flutter — Dart 3+, Riverpod 2, GoRouter, cross-platform targets.

The only dedicated TUI design-and-craft hierarchy in the public LLM-coding ecosystem — architect decides composition, designer crafts each screen, implementer writes View/Update code.

  • fft-tui-architect — strategic composition, Charm-ecosystem adoption, layout budget, focus model, and the visual full-screen QA gate. Dispatch first for composition questions.
  • fft-tui-designer — per-screen visual design: the mockup, visual hierarchy, color/glyph/label co-render, spacing-token application, state tables, and the “gorgeous” bar. Hands fft-tui an implementation-ready screen spec.
  • fft-tui — Charm.sh BubbleTea, Elm MVU, lipgloss, 60 fps render budgets, CVD-safe accessibility.

The gatekeepers of production readiness.

  • fft-qa — quality assurance strategy, test automation, and regression gates.
  • fft-code-reviewer — standards enforcement, pattern analysis, architectural validation.
  • fft-security — threat modeling, OWASP compliance, defense-in-depth.
  • fft-web-quality — Playwright 4-viewport screenshots, Lighthouse CWV, axe-core WCAG AA, broken-link detection.

Ship it, run it, observe it.

  • fft-devops-agent — CI/CD, Docker, Kubernetes, Terraform, monitoring, cost optimization.
  • fft-github — git workflow, PR automation, issue management, release operations.
  • fft-documentation — technical writing, API docs, architecture docs, user guides.
  • fft-documenter-br — Brazilian Portuguese documentation with correct diacritics.
  • fft-documenter-es — Spanish documentation: RAE orthography, tildes, inverted punctuation, Latin-American register.

Hierarchical: start with fft-ml-architect for strategy, then route to a specialist.

  • fft-ml-architect — strategic advisor: model selection, RAG vs fine-tuning, GPU allocation.
  • fft-llm-openweight — local LLM deployment: vLLM, Ollama, quantization, Qwen/Llama/DeepSeek.
  • fft-rag-engineer — retrieval architecture, pgvector, embeddings, chunking, reranking.
  • fft-agent-frameworks — CrewAI, LlamaIndex, LangChain, custom multi-agent orchestration.
  • fft-performance — profiling, caching, load testing, scaling strategies.
  • fft-medical — Brazilian healthcare compliance (ANS, TISS, TUSS, CBHPM, LGPD).
  • fft-agent-creator — creates and validates new FlowForge agents from specifications.

For the complete side-by-side list, see the catalog index. To understand how the Maestro selects and coordinates these agents, read Core Concepts: Maestro Pattern.