README

A structured multi-agent implementation pipeline for Claude Code. Spawns specialized agents (planner, researchers, implementers,…

HN3K 3 updated 1mo ago
Claude CodeGeneric
View source ↗
# GOAT — Agent Team Pipeline

A structured multi-agent implementation pipeline for Claude Code. Spawns specialized agents (planner, researchers, implementers, reviewers) through a 6-phase workflow with built-in quality gates.

## Prerequisites

- **Claude Code** CLI with agent team support (`TeamCreate`, `Agent`, `SendMessage`, `TaskCreate`)
- **codebase-index-tools** CLI installed at repo root (Python: `python -m codebase_index_tools`, Node: `node codebase-index-tools/cli.js`)
- **Codebase-Index/** (or equivalent) directory with index mappings for your project

## Setup (GOAT-CEO context)

This repo (`GOAT-CEO`) is the meta-orchestration repo — it contains skill definitions, specs, and pipeline logs. It does not contain runtime application code. The GOAT pipeline is already configured and ready to use.

1. The `.claude/commands/goat-team/` directory is already set up in this repo — no copying needed.

2. Custom agent definitions live in `.claude/agents/`:
   - `team-architect.md` — Planner/architect roles (Phases 1, 2)
   - `team-researcher.md` — Codebase and technical researchers (Phase 2)
   - `team-implementer.md` — Implementers and index updater (Phases 3, 4)
   - `team-verifier.md` — Reviewers (Phase 5)

3. The `goat.md` command is already configured with GOAT-CEO project-specific rules. If you use this pipeline in another repo, copy the `.md` files and update the **Project-Specific Rules** section in `goat.md`:

   ```markdown
   ## Project-Specific Rules (include in agent prompts when relevant)

   - Your build system and test commands
   - Code patterns and conventions
   - Architecture boundaries
   - Any repo-specific constraints
  1. Check CLAUDE.md in the target repo for the correct tooling invocation before running agents that use codebase-index-tools.

Usage

Full pipeline (plan → research → implement → review):

/goat Fix the authentication bug in the login flow

Plan only (stops before implementation):

/goat-plan Design the new caching layer

Review only (assumes implementation is already done):

/goat-review Review the changes in agent-workspace/

Index maintenance:

/index-check

Files

File Purpose
goat.md Main entry point — overseer orchestrates all phases
goat-plan.md Plan-only variant (Phases 1-2, stops before implementation)
goat-review.md Review-only variant (assumes implementation complete)
planner.md Planner agent role — creates/revises plans and manifests
codebase-researcher.md Codebase researcher — finds upstream/downstream risks
technical-researcher.md Technical researcher — assesses approach quality
implementer.md Implementer agent — executes one manifest batch
index-updater.md Index updater — content-aware index accuracy layer
reviewer.md Reviewer agent — independent verification
index-check.md Standalone index audit and update utility

Agent Types

The pipeline uses custom agent definitions from .claude/agents/:

Agent type Used by roles
team-architect planner, planner-review, planner-manifest
team-researcher codebase-researcher, tech-researcher
team-implementer implementer-{N}, index-updater
team-verifier reviewer-a, reviewer-b

Pipeline Phases

  1. Planning — Planner loads index context, creates PLAN.md and shared index-context.md
  2. Research & Revision Loop — Two researchers investigate in parallel, planner revises, loop until clean. On exit, planner generates IMPLEMENTATION-MANIFEST.md inline
  3. Implementation — Implementers execute batches (parallel when no file conflicts)
  4. Index Update — Index updater ensures indexes match code changes + progressive enrichment
  5. Review — Two independent reviewers verify implementation and Index Updater completeness
  6. Finalize — Overseer evaluates verdicts, handles failures, summarizes

Artifacts

All pipeline artifacts are written to agent-workspace/ at the repo root:

File Owner Purpose
PLAN.md Planner Implementation plan
RESEARCH-LOG.md Researchers Running log of findings and signals
ISSUE-TRACKER.md All agents Issues with severity tracking
IMPLEMENTATION-MANIFEST.md Planner Batched implementation tasks
REVIEW-LOG.md Implementers, Index Updater, Reviewers Index checks and review verdicts

Customization

The only file that needs per-repo customization is goat.md. Everything else is generic and works with any repo that has codebase-index-tools and a Codebase-Index/ (or equivalent) directory set up. ```

Maintain README?

Let people know it's listed here — add the badge (live metrics, light/dark aware) or a plain link to your README or docs.

[README on getagentictools](https://getagentictools.com/loops/hn3k-goat-agent-team-pipeline?ref=badge)