fenix-plugin — AI Agent Team

fenix-plugin turns your AI coding agent into a full product team. It ships as a Claude Code plugin (and as installable skills for Cursor, Codex, and Gemini), bringing 5 specialized agents and 10 workflow skills to your development workflow — with OAuth auth and automatic cross-session memory.

Installation

Claude Code

/plugin marketplace add fenix-assistant/fenix-plugin

An OAuth flow opens in your browser. Sign in with your Fenix account. No PAT or environment variables needed.

Cursor / OpenAI Codex / Gemini CLI

1. Go to plugin or extension settings
2. Add plugin via URL (available in Fenix → Settings → Integrations)
3. Authenticate via the OAuth popup

The 5 agents

Fenix Orchestrator

Runs at the start of every session. Loads workspace context, searches memories for session state and recent decisions, detects the current workflow phase, and routes to the appropriate agent.

Fenix Planner

Two modes:

Product planning — creates epics, features, and stories from natural language descriptions with acceptance criteria.

Technical planning — breaks stories into tasks. Always identifies ambiguities and presents trade-off options before creating tasks. Locks decisions as persistent memories. Never writes code.

Fenix Builder

Implements tasks with TDD discipline:

• Reads all relevant files before writing any code
• Writes failing tests first (Red → Green → Refactor)
• Executes in waves — complex changes split across commits
• Updates task status as work progresses
• Will not start without Planner-created tasks

Fenix Reviewer

Reviews implementation against:

• Story acceptance criteria
• Architectural decisions from memories
• Code quality: naming, complexity, test coverage, security

Fenix Deliverer

Handles story closure:

• Marks tasks and story done
• Saves a delivery summary as a memory
• Creates or updates linked documentation
• Cleans up TODO comments and temp code

The 10 skills

Skills are reusable instruction sets — exportable and usable in any agent.

Export a skill

Action: skill_export_merged
query: "fenix-technical-development"

Returns the full skill content — paste into any agent's system prompt.

Example workflow

you: "implement story S-247: add rate limiting to auth endpoint"

orchestrator → loads context, searches memories for S-247 and auth decisions
planner → confirms tasks exist (T-891, T-892, T-893)
builder → reads auth module, writes failing test for rate limit
builder → implements TokenBucketGuard, tests pass
builder → updates T-891 to done, moves to T-892
reviewer → checks against acceptance criteria
deliverer → closes S-247, saves delivery memory, updates auth docs

Memory protocol

Agents save context automatically:

Workflow gates

These rules are non-negotiable:

1. No implementation without tasks — Builder refuses to write code without a task list
2. No skipping discussion phase — Planner always presents options before creating tasks
3. No status changes without work — Deliverer won't close a story unless all tasks are done
4. TDD when applicable — if you can write expect(fn(input)).toBe(output) before writing fn, you must