Autonomous development pipeline using Claude Code Agent Teams. Orchestrates brainstorm, plan, implement, verify, reflect cycles with specialized teammates and GPT-5.2/5.3-Codex integration.
Use when starting a new project or when a track needs ideation and product shaping. Human-in-loop only. <example> Context: User wants to explore ideas for a new feature user: "Let's brainstorm the authentication system" assistant: "Spawning brainstormer for BMAD-style ideation" <commentary>New feature needs facilitated exploration before planning</commentary> </example>
Use when a TASK packet is ready for implementation. Implements exactly what the packet says. <example> Context: Task packet ready for coding user: "Implement task auth-P1-T02-jwt" assistant: "Dispatching to GPT-5.3-Codex for implementation" <commentary>Atomic task execution via Codex</commentary> </example>
Use at phase or track boundaries to evaluate drift, quality of direction, and whether tasks remain correct. <example> Context: Track complete, need boundary evaluation user: "Evaluate auth track completion" assistant: "Running boundary evaluation: drift, spec alignment, direction" <commentary>Phase boundary requires meta-evaluation</commentary> </example>
Use when a repo is detected as brownfield and we need a one-shot discovery pass before brainstorm. <example> Context: Lead ran detect and got brownfield user: "Run discovery for this repo" assistant: "Spawning discoverer to map codebase evidence into docs/discovery.md" <commentary>Brownfield discovery needed before ideation starts</commentary> </example>
Use only at track boundary to propose living-doc reconciliation after Conductor emits a trigger. <example> Context: Track completed with CONTINUE decision user: "Run doc reconciliation for v32-realignment" assistant: "Preparing track-boundary reconciliation proposals from diff and living docs" <commentary>Doc-keeper now runs only at track boundaries</commentary> </example>
Conductor verdict format, drift protocol, boundary evaluation.
Core deadfish invariants and protocols. Referenced by all teammates.
Brownfield discovery protocol (detect, collect evidence, analyze, and write docs/discovery.md).
Track-boundary living-doc reconciliation and debate protocol.
Implementation constraints, git conventions, Codex MCP usage.
Modifies files
Hook triggers on file write and edit operations
Uses power tools
Uses Bash, Write, or Edit tools
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
"Only a dead fish follows the flow."
An opinionated, multi-model workflow for building real software with AI agent teams.
Combining the best of BMAD, Oh My OpenCode, Google Conductor, and GSD into one pipeline.
This is my attempt at creating what I would consider an efficient and optimized workflow to help you develop any — or at least most — ideas you might have.
Over the last couple of months I've had the chance to play around and get great results from some amazing open-source projects. Each one taught me something different, and each one left me wanting just a little more.
I tried to combine the different strengths of each and improve from the combination of all the concepts together. The result is deadfish-teams — built on top of Claude Code Agent Teams, Anthropic's brand-new multi-agent system that shipped with Opus 4.6 on February 5, 2026. Agent Teams lets multiple Claude Code instances work in parallel as a coordinated team — not subagents locked to one context, but fully independent sessions that communicate directly with each other. Deadfish gives that raw capability a purpose: a structured pipeline where each teammate has a defined role, strict boundaries, and a protocol to follow.
This is day-one tooling for a day-one feature. We're pushing the envelope together. 🚀
| 🧠 |
BMAD Method — the brainstorming engineI fell in love with BMAD's brainstorming capabilities. Not the "give me 10 ideas" kind — this is world-class, structured ideation that actually converges into real requirements. Version 6 is just crazy good. Deadfish borrows BMAD's multi-template brainstorming pipeline (7 templates that guide you from wild ideas to a concrete spec with acceptance criteria). But here's the thing. BMAD plans everything upfront. All phases, all tasks, all at once. And 82 tasks later, you're still following the same rigid structure — even though at task 32 you realized a better design was possible. But it was already planned, so... you keep going. 🤷 Deadfish keeps BMAD's brainstorming brilliance but replaces the rigid execution with something more adaptive. |
| 🔁 |
Oh My OpenCode — the relentless loopThis one is the looping method on steroids. It auto-answers prompts to keep things moving, and the results were genuinely impressive — I had a lot of success with it. What it showed me was how far you can push a CLI tool with extensions, hooks, and a composable plugin architecture. Deadfish takes that plugin DNA: it installs as a Claude Code plugin with hooks, skills, and agent definitions that you can swap, extend, and upgrade without losing your customizations. ♻️ |
| 🧭 |
Google Conductor — the adaptive navigator |
npx claudepluginhub fredasterehub/deadfish-teamsMulti-modal software creation pipeline with milestone-scoped lifecycle and dynamic worker cycling. From idea to deployed code in 7 autonomous steps.
Production-ready Claude Code configuration with role-based workflows (PM→Lead→Designer→Dev→QA), safety hooks, 44 commands, 19 skills, 8 agents, 43 rules, 30 hook scripts across 19 events, auto-learning pipeline, hook profiles, and multi-language coding standards
Structured project planning and execution through brainstorm, spec, and build phases across three execution tiers: sequential, delegated sub-agents, and full agent teams
AI-powered development tools for code review, research, design, and workflow automation.
Describe your goal, approve the spec, then step away — Claude and Codex loop together until it's right.
Multi-agent software evolution. Autonomously study, strategize, build, review, and evaluate code changes using specialist AI agents.
Plugin for effective agentic development