The Agent Catalog

FlowForge doesn’t treat AI as a single genie. It runs a bench of specialists — each one a deeply-prompted expert in one discipline, each dispatched by ACE for exactly the work it’s best at. This chapter is the roster: 36 specialist agents, split by discipline, one entry each.

The same catalog is published on the web at flowforgesoft.com/agents — this chapter mirrors that structure so the manual and the site read as one product.

Routing, in one sentence: reach for the specialist when it multiplies you — parallel work, deep domain expertise, or an independent review — and say why when you don’t. Some disciplines are hierarchical: a strategy-level architect decides the approach and hands a brief to a leaf agent that implements it. Those pairs are noted below.

---

Project Ingest & Discovery

fft-code-explorer — The phase-0 project-ingest specialist. Reads the source tree, the database schema (inferred from migrations — no live credentials needed), configs, and lockfiles to detect what a project is: frameworks and exact versions, languages, dependencies, architecture, entry points. It never guesses — it verifies at source — and then scopes the downstream specialist wave. This is where landing a legacy codebase begins.

---

Architecture & Planning

fft-requirements-analyst — The anti-fabrication interrogator. Grills scope before any architecture is drawn, captures the customer’s intent verbatim, and produces the structured requirements packet everyone downstream builds from. Read-only: it asks and reports; it does not write the code.

fft-architecture — The system architect. Owns structural decisions, design patterns, and technical leadership — the SOLID, do-it-right-and-do-once decisions that are expensive to undo.

fft-project-manager — Breaks features into tasks, builds the milestone and ticket structure, and keeps the plan organized.

fft-product-owner — Prioritizes for value, aligns stakeholders, and guards the line between what’s in scope and what isn’t.

---

API & Backend

fft-api-designer — REST, GraphQL, gRPC, and WebSocket API design: contracts, OpenAPI specs, versioning, and developer-friendly surfaces.

fft-backend — Server-side implementation across Node.js, Python, Ruby, Java, and Go — auth, services, and scalable architectures.

fft-database — Data modeling, schema design, and query optimization — the ACID-minded data layer.

---

Frontend & Design

Design work is hierarchical: the design authority decides which guidelines govern this surface and hands a brief to the right leaf implementer.

fft-design-architect — The master design authority — the design analog of the system architect. Holds the guideline corpus (Apple HIG, Material 3, WCAG, the FlowForge design canon and brand manual) and emits the per-surface design brief that the leaf designers implement against and the reviewer judges GOLDEN against. Design strategy starts here.

fft-frontend — React, Vue, Angular, and modern UI — the web implementation leaf: component architecture, animation, and responsive, accessible interfaces.

fft-designer — The UI/UX and design-system leaf for web and print/papelaria surfaces (reports, PDFs) — user research, component systems, and accessibility, implemented against the brief.

fft-brand-architect — Brand identity, color theory, typography systems, and design tokens — the brand leaf.

---

Mobile

fft-ios — Native iOS: Swift, SwiftUI, UIKit, structured concurrency, and the App Store lifecycle.

fft-android — Native Android: Kotlin, Jetpack Compose, Material Design 3, and Play Store compliance.

fft-flutter — Cross-platform Dart 3+ with Riverpod and GoRouter, adaptive UI across iOS/Android/Web/Desktop.

---

Terminal UI

Terminal-UI work is hierarchical: an architect decides composition, a designer crafts each screen’s visual, and a leaf writes the view code — architect → designer → tui.

fft-tui-architect — Strategic terminal-UI composition: full-screen layout, focus across panes, responsive breakpoints, Charm-ecosystem-adoption discipline, and the visual QA gate. Decides composition; delegates per-screen visual design to fft-tui-designer and line-code to fft-tui.

fft-tui-designer — The terminal visual-design leaf: per-screen mockups, visual hierarchy, color/glyph/label co-render, spacing-token application, and state tables — the “gorgeous” bar. Holds the founder’s taste as canon (the TUI Design Manual, the Spacing Canon, the header/frame blueprint) and hands fft-tui an implementation-ready screen spec.

fft-tui — Deep Charm.sh craft: BubbleTea, lipgloss, bubbles, glamour, and the Elm Model-View-Update pattern, with accessibility, responsive design, and a 60fps budget. The cockpit you use is its work.

---

Quality & Security

fft-qa — Quality assurance and test strategy: TDD, coverage, and testing across unit, integration, E2E, and performance.

fft-code-reviewer — The code-quality guardian and the sole arbiter of the GOLDEN verdict. It verifies a worker’s claims against the primary sources, not against the worker’s own word.

fft-security — Threat modeling, OWASP compliance, and defense-in-depth security review.

fft-web-quality — Empirical web-PR validation: multi-viewport screenshots, Core Web Vitals, axe-core WCAG scans, broken-link detection — evidence, not opinion.

---

Operations

fft-devops-agent — CI/CD, Kubernetes, Docker, Terraform, monitoring, and cost optimization.

fft-github — Git and GitHub workflows: branches, PRs, issue mechanics, and repository management.

---

Documentation

fft-documentation — Technical writing, API docs, and knowledge management — comprehensive, maintainable documentation.

fft-documenter-br — Brazilian-Portuguese documentation, with rigorous PT-BR grammar and diacritics. It rejects European Portuguese; all PT-BR content routes through it.

fft-documenter-es — Spanish documentation, with RAE orthography, tildes and inverted punctuation, and a Latin-American register. The Spanish counterpart to fft-documenter-br.

---

Marketing & Content

fft-content-strategist — Landing-page copy, SEO, email, micro-copy, and conversion — the brand-voice architecture.

fft-social-media — Multi-platform social content, calendar, engagement, and analytics.

---

ML / AI

ML/AI work is hierarchical: a strategist decides the approach, then delegates to the right implementation specialist.

fft-ml-architect — Strategic ML/AI guidance: model selection, RAG-vs-fine-tuning, and GPU allocation. Decides; delegates to the specialists below.

fft-llm-openweight — Local and on-premise open-weight models (Llama, Mistral, Qwen, DeepSeek), quantization, and serving with vLLM / Ollama / llama.cpp.

fft-rag-engineer — Retrieval architecture: chunking, pgvector, embeddings, reranking, and RAG evaluation.

fft-agent-frameworks — Agentic applications with CrewAI, LlamaIndex, LangChain, AutoGen, and custom multi-agent systems.

---

Performance

fft-performance — Full-stack performance: profiling, load testing, caching, query optimization, and monitoring.

---

Medical

fft-medical — Brazilian healthcare: ANS, TISS, TUSS, and CBHPM, medical document processing, and LGPD/HIPAA compliance.

---

Meta

fft-agent-creator — The meta-agent: it creates and validates new FlowForge specialists, so the bench itself can grow to fit a project.

---

Why it matters

Specialists are how FlowForge scales without losing depth. You describe intent; ACE routes the work to the expert bench, dispatches an independent reviewer to check it, and keeps the whole thing pointed at what you actually asked for. The bench is not fixed — fft-agent-creator mints new specialists as a project needs them — which is why the roster on flowforgesoft.com/agents and the one installed on your machine are living lists, not a frozen brochure.