{
  "id": "page:library-metaswarm",
  "_kind": "Page",
  "_file": "wiki/library/metaswarm.md",
  "_cluster": "wiki",
  "attributes": {
    "nodeKind": "Page",
    "title": "Metaswarm Methodology (Library)",
    "displayName": "Metaswarm Methodology (Library)",
    "slug": "library/metaswarm",
    "articlePath": "wiki/library/metaswarm.md",
    "article": "\n# Metaswarm Methodology\n\n**Source**: [dsifry/metaswarm](https://github.com/dsifry/metaswarm) by David Sifry\n**Category**: Autonomous Multi-Agent Orchestration / Issue-to-PR Lifecycle\n**License**: See upstream repository\n\n## Overview\n\nMetaswarm is an autonomous multi-agent orchestration framework that manages the complete lifecycle from GitHub issue to merged pull request. It coordinates 12 specialized agents through a rigorous 7-phase workflow with quality gates as blocking state transitions, adversarial reviews with fresh reviewers, and knowledge persistence across sessions.\n\n## Core Principles\n\n- **Trust Nothing, Verify Everything, Review Adversarially** - Quality gates are blocking, never advisory\n- **TDD Mandatory** - Write tests first, watch them fail, then implement (100% coverage targets)\n- **Fresh Reviewer Rule** - On re-review after FAIL, spawn new reviewer with no memory (prevents anchoring bias)\n- **Independent Validation** - Orchestrator runs quality gates directly, never trusts subagent self-reports\n- **Human Checkpoints** - Planned pauses at critical boundaries (schema changes, security code, new patterns)\n- **Knowledge Persistence** - Extract learnings while context is fresh, persist for cross-session continuity\n\n## Process Files\n\n| Process | File | Description | Task Count |\n|---------|------|-------------|------------|\n| Issue Orchestrator | `metaswarm-orchestrator.js` | Full 7-phase lifecycle: research to PR | 12 |\n| Design Review Gate | `metaswarm-design-review.js` | 6 parallel specialist reviews, unanimous approval | 7 |\n| Execution Loop | `metaswarm-execution-loop.js` | 4-phase cycle: Implement -> Validate -> Review -> Commit | 4 |\n| Swarm Coordinator | `metaswarm-swarm-coordinator.js` | Multi-issue parallel management across worktrees | 6 |\n| Knowledge Cycle | `metaswarm-knowledge-cycle.js` | Context priming and self-reflection | 5 |\n| PR Shepherd | `metaswarm-pr-shepherd.js` | PR lifecycle through merge | 3 |\n\n## Skills Catalog\n\n| Skill | Directory | Description |\n|-------|-----------|-------------|\n| orchestrated-execution | `skills/orchestrated-execution/` | 4-phase execution loop with quality gates |\n| design-review-gate | `skills/design-review-gate/` | 6-agent parallel design review |\n| plan-review-gate | `skills/plan-review-gate/` | 3 adversarial plan reviewers |\n| adversarial-review | `skills/adversarial-review/` | Fresh reviewer with binary PASS/FAIL |\n| knowledge-curation | `skills/knowledge-curation/` | Context priming and self-reflection |\n| work-unit-decomposition | `skills/work-unit-decomposition/` | DoD items, file scope, dependencies |\n| pr-shepherding | `skills/pr-shepherding/` | PR lifecycle management through merge |\n| external-tool-coordination | `skills/external-tool-coordination/` | Cross-model AI tool integration |\n\n## Agents Catalog\n\n| Agent | Directory | Role |\n|-------|-----------|------|\n| issue-orchestrator | `agents/issue-orchestrator/` | Master coordinator per issue |\n| researcher | `agents/researcher/` | Codebase exploration and analysis |\n| architect | `agents/architect/` | Implementation planning and decomposition |\n| product-manager | `agents/product-manager/` | Use case and scope validation |\n| designer | `agents/designer/` | UX/API design review |\n| security-design | `agents/security-design/` | Threat modeling and OWASP analysis |\n| cto | `agents/cto/` | TDD readiness and codebase alignment |\n| coder | `agents/coder/` | TDD implementation specialist |\n| code-reviewer | `agents/code-reviewer/` | Fresh adversarial reviewer |\n| security-auditor | `agents/security-auditor/` | Implementation security review |\n| pr-shepherd | `agents/pr-shepherd/` | PR lifecycle management |\n| swarm-coordinator | `agents/swarm-coordinator/` | Multi-issue parallel orchestration |\n\n## Workflow Lifecycle\n\n```\nIssue -> Research -> Plan -> Plan Review Gate (3 adversarial) -> Preflight -> Design Review Gate (6 parallel, unanimous) -> Work Unit Decomposition -> Execution Loop (Implement -> Validate -> Adversarial Review -> Commit) x N -> Final Review -> Self-Reflect -> PR -> Shepherd -> Merge\n```\n\n## Quality Gates (Blocking)\n\n1. **Plan Review Gate** - 3 adversarial reviewers (Feasibility, Completeness, Scope)\n2. **Design Review Gate** - 6 parallel specialists (ALL must approve, max 3 iterations)\n3. **Quality Gate Validation** - Independent tsc, eslint, vitest (blocking, never advisory)\n4. **Adversarial Code Review** - Fresh reviewer, binary PASS/FAIL (max 3 attempts -> escalate)\n5. **Coverage Gate** - 100% target across lines/branches/functions/statements\n\n## Anti-Patterns (Enforced)\n\n- Self-certifying (trusting subagent claims)\n- Combining execution phases into single steps\n- Reusing reviewers after FAIL verdict\n- Passing previous review findings to new reviewers\n- Treating quality gate failures as advisory\n- Proceeding past human checkpoints without explicit approval\n- Using --no-verify on commits or force-pushing\n\n## Philosophy\n\n- **Adversarial over collaborative** review for verified compliance\n- **Independent validation** over trusted reporting\n- **Blocking gates** over advisory recommendations\n- **Fresh reviewers** over experienced reviewers (prevent bias)\n- **Knowledge extraction** before context is lost\n- **Human escalation** after bounded retry (3 attempts)\n",
    "documents": [
      "specialization:metaswarm"
    ]
  },
  "outgoingEdges": [
    {
      "from": "page:library-metaswarm",
      "to": "specialization:metaswarm",
      "kind": "documents"
    }
  ],
  "incomingEdges": [
    {
      "from": "page:index",
      "to": "page:library-metaswarm",
      "kind": "contains_page"
    }
  ]
}