Claude Code Plugins

Community-maintained marketplace

Feedback

Reference for parallel agent deployment protocols (search-party, qa-party, plan-party)

Install Skill

1Download skill
2Enable skills in Claude

Open claude.ai/settings/capabilities and find the "Skills" section

3Upload to Claude

Click "Upload skill" and select the downloaded ZIP file

Note: Please verify skill by going through its instructions before using it.

SKILL.md

name parties
description Reference for parallel agent deployment protocols (search-party, qa-party, plan-party)

Party Protocols

Purpose: Display the PAI party protocol structures Trigger: /parties Updated: 2026-01-03


Party Overview

Party Staff Probes/Agents Purpose Invocation
SEARCH_PARTY G-2 (Intel) 120 probes Reconnaissance /search-party
QA_PARTY IG (Inspector General) 120 agents Validation /qa-party
PLAN_PARTY G-5 (Strategic) 10 probes Planning /plan-party

SEARCH_PARTY (G-2 Reconnaissance)

G2_RECON (Commander)
    │
    ├─ G2-BACKEND-CORE ────┬─ PERCEPTION ─── Surface state, logs, errors
    │  (backend/app/api/)  ├─ INVESTIGATION  Dependencies, call chains
    │                      ├─ ARCANA ─────── ACGME rules, security
    │                      ├─ HISTORY ────── Git log, recent changes
    │                      ├─ INSIGHT ────── Design decisions, intent
    │                      ├─ RELIGION ───── CLAUDE.md compliance
    │                      ├─ NATURE ─────── Over-engineering detection
    │                      ├─ MEDICINE ───── System diagnostics, leaks
    │                      ├─ SURVIVAL ───── Edge cases, failure modes
    │                      └─ STEALTH ────── Hidden coupling, blind spots
    │
    ├─ G2-BACKEND-MODELS ──┬─ (10 probes)
    │  (backend/app/models/)
    │
    ├─ G2-SCHEDULING ──────┬─ (10 probes)
    │  (backend/app/scheduling/)
    │
    ├─ G2-RESILIENCE ──────┬─ (10 probes)
    │  (backend/app/resilience/)
    │
    ├─ G2-FRONTEND-CORE ───┬─ (10 probes)
    │  (frontend/src/components/)
    │
    ├─ G2-FRONTEND-HOOKS ──┬─ (10 probes)
    │  (frontend/src/hooks/)
    │
    ├─ G2-TESTS-BACKEND ───┬─ (10 probes)
    │  (backend/tests/)
    │
    ├─ G2-TESTS-FRONTEND ──┬─ (10 probes)
    │  (frontend/__tests__/)
    │
    ├─ G2-MCP ─────────────┬─ (10 probes)
    │  (mcp-server/)
    │
    ├─ G2-DOCS ────────────┬─ (10 probes)
    │  (docs/, .claude/)
    │
    ├─ G2-INFRASTRUCTURE ──┬─ (10 probes)
    │  (docker, .github/, scripts/)
    │
    └─ G2-SECURITY ────────┬─ (10 probes)
       (backend/app/core/)

Total: 12 G-2 teams × 10 probes = 120 probes

D&D Probe Lenses

Probe Lens What It Finds
PERCEPTION Surface state Logs, errors, health checks
INVESTIGATION Connections Dependencies, imports, call chains
ARCANA Domain knowledge ACGME rules, resilience patterns
HISTORY What changed Git log, recent PRs, migrations
INSIGHT Intent/design Why built this way, tech debt
RELIGION Sacred law CLAUDE.md compliance, heresies
NATURE Organic growth Over-engineering, forced patterns
MEDICINE Diagnostics Unhealthy components, leaks
SURVIVAL Edge resilience Brittleness, failure modes
STEALTH Hidden elements Hidden coupling, blind spots

QA_PARTY (IG Validation)

COORD_QUALITY (Commander)
    │
    ├─ QA-UNIT-TESTS ──────┬─ pytest backend/tests/unit/
    │                      ├─ Coverage analysis
    │                      ├─ Mock validation
    │                      └─ Edge case detection
    │
    ├─ QA-INTEGRATION ─────┬─ pytest backend/tests/integration/
    │                      ├─ API endpoint validation
    │                      ├─ Database transaction tests
    │                      └─ Service interaction tests
    │
    ├─ QA-FRONTEND ────────┬─ npm test (Jest)
    │                      ├─ Component rendering
    │                      ├─ Hook behavior
    │                      └─ State management
    │
    ├─ QA-E2E ─────────────┬─ Playwright tests
    │                      ├─ Critical user journeys
    │                      ├─ Cross-browser validation
    │                      └─ Accessibility checks
    │
    ├─ QA-LINT-BACKEND ────┬─ ruff check
    │                      ├─ ruff format --check
    │                      └─ mypy type checking
    │
    ├─ QA-LINT-FRONTEND ───┬─ npm run lint
    │                      ├─ npm run type-check
    │                      └─ Prettier validation
    │
    ├─ QA-BUILD ───────────┬─ npm run build
    │                      ├─ Docker build validation
    │                      └─ Asset optimization
    │
    ├─ QA-SECURITY ────────┬─ Dependency audit
    │                      ├─ Secret scanning
    │                      ├─ SAST analysis
    │                      └─ OWASP checks
    │
    ├─ QA-ACGME ───────────┬─ Compliance validation
    │                      ├─ 80-hour rule tests
    │                      ├─ 1-in-7 rule tests
    │                      └─ Supervision ratio tests
    │
    ├─ QA-MIGRATIONS ──────┬─ alembic upgrade/downgrade
    │                      ├─ Schema validation
    │                      └─ Data integrity checks
    │
    ├─ QA-API-CONTRACT ────┬─ OpenAPI validation
    │                      ├─ Schema drift detection
    │                      └─ Breaking change detection
    │
    └─ QA-PERFORMANCE ─────┬─ Load test baselines
                           ├─ Memory profiling
                           └─ Query optimization

Total: 12 QA teams × 10 validators = 120 agents


PLAN_PARTY (G-5 Strategic)

G5_PLANNING (Commander)
    │
    ├─ PROBE-ARCHITECTURE ─── System design implications
    │
    ├─ PROBE-DEPENDENCIES ─── Library/service dependencies
    │
    ├─ PROBE-MIGRATION ────── Database schema changes
    │
    ├─ PROBE-TESTING ──────── Test strategy requirements
    │
    ├─ PROBE-SECURITY ─────── Security considerations
    │
    ├─ PROBE-PERFORMANCE ──── Performance implications
    │
    ├─ PROBE-ACGME ────────── Compliance impact
    │
    ├─ PROBE-FRONTEND ─────── UI/UX requirements
    │
    ├─ PROBE-RESILIENCE ───── Failure mode analysis
    │
    └─ PROBE-ROLLBACK ─────── Reversion strategy

Total: 10 planning probes

Plan Output Structure

Each probe contributes to:

  1. Implementation Steps - Ordered task breakdown
  2. Critical Files - Files requiring modification
  3. Risk Assessment - Potential issues and mitigations
  4. Dependencies - Prerequisites and blockers
  5. Validation Criteria - How to verify success

Economics: Zero Marginal Wall-Clock Cost

Sequential (BAD):        Parallel (GOOD):
120 probes × 30s each    120 probes × 30s in parallel
Total: 3600s (1 hour)    Total: 30s (120x faster)

All probes run in parallel. No cost savings from running fewer.


IDE Crash Prevention

CRITICAL: Never spawn 8+ agents directly from ORCHESTRATOR

CORRECT:                           WRONG:
ORCHESTRATOR → 1 Commander         ORCHESTRATOR → 12 G-2s directly
                ↓                                  ↓
         Commander manages N               IDE CRASH

Always route through the party Commander (G2_RECON, COORD_QUALITY, G5_PLANNING).


Quick Reference

Command Staff Scale Use For
/search-party G-2 120 probes Codebase exploration
/qa-party IG 120 agents Comprehensive validation
/plan-party G-5 10 probes Implementation planning

Run /parties anytime for this reference.