SEARCH_PARTY Skill

Purpose: Coordinated parallel reconnaissance with 120 specialized probes Created: 2025-12-31 Trigger: /search-party command Aliases: /recon, /search, /sp

---

When to Use

Deploy SEARCH_PARTY when you need comprehensive codebase intelligence:

• Investigating unfamiliar code areas
• Pre-task reconnaissance for complex changes
• Bug investigation with unclear root cause
• Security or compliance reviews
• Technical debt assessment
• Architecture review

Do NOT use for:

• Simple, well-understood tasks
• Single-file changes with known impact
• When only one lens is needed

---

Economics: Zero Marginal Wall-Clock Cost

Critical Understanding: Parallel agents with the same timeout cost nothing extra in wall-clock time.

Sequential (BAD):        Parallel (GOOD):
10 probes × 30s each     10 probes × 30s in parallel
Total: 300s              Total: 30s (10x faster)

Implication: Always spawn all probes. There is no cost savings from running fewer.

---

Deployment Pattern

Standard Deployment: 12 G-2s × 10 Probes = 120 Probes

Deploy 12 G-2 RECON agents in parallel, each commanding a party of 10 probes:

G-2 Mission	Target Domain	10 Probes
G2-BACKEND-CORE	backend/app/api/, backend/app/services/	Full party
G2-BACKEND-MODELS	backend/app/models/, backend/alembic/	Full party
G2-SCHEDULING	backend/app/scheduling/	Full party
G2-RESILIENCE	backend/app/resilience/	Full party
G2-FRONTEND-CORE	frontend/src/components/, frontend/src/app/	Full party
G2-FRONTEND-HOOKS	frontend/src/hooks/, frontend/src/lib/	Full party
G2-TESTS-BACKEND	backend/tests/	Full party
G2-TESTS-FRONTEND	frontend/__tests__/	Full party
G2-MCP	mcp-server/	Full party
G2-DOCS	docs/, .claude/	Full party
G2-INFRASTRUCTURE	docker-compose.yml, .github/, scripts/	Full party
G2-SECURITY	backend/app/core/, backend/app/api/deps.py	Full party

Probe Composition (Per Party)

Each G-2 deploys these 10 D&D-inspired probes in parallel:

Probe	Lens	D&D Analog	What It Finds
PERCEPTION	Surface state	Spot check	Logs, errors, health checks, what's immediately visible
INVESTIGATION	Connections	Search check	Dependencies, imports, call chains, why things connect
ARCANA	Domain knowledge	Arcana check	ACGME rules, resilience patterns, security implications
HISTORY	What changed	History check	Git log, recent PRs, migrations, blame
INSIGHT	Intent/design	Insight check	Why built this way, design decisions, tech debt
RELIGION	Sacred law	Religion check	CLAUDE.md compliance, pattern heresies, rituals
NATURE	Organic growth	Nature check	Over-engineering, natural vs forced patterns
MEDICINE	System diagnostics	Medicine check	Unhealthy components, resource exhaustion, leaks
SURVIVAL	Edge resilience	Survival check	Brittleness, missing error handling, failure modes
STEALTH	Hidden elements	Stealth check	Hidden coupling, security blind spots, invisible state

---

Invocation

Full Deployment (120 probes)

/search-party

Deploys all 12 G-2s with full probe parties.

Targeted Deployment (10 probes)

/search-party backend/app/scheduling/

Deploys single G-2 on specific target.

Quick Search (focused)

/search-party --quick backend/

Deploys 6 critical probes (PERCEPTION, INVESTIGATION, ARCANA, HISTORY, INSIGHT, STEALTH).

---

IDE Crash Prevention (CRITICAL)

DO NOT have ORCHESTRATOR spawn 12 G-2 agents directly. This causes IDE seizure and crashes.

CORRECT Pattern:

ORCHESTRATOR → spawns 1 G2_RECON (G-2 Commander)
                    ↓
              G2_RECON deploys 12 G-2 teams internally
              (manages parallelism, synthesizes results)

WRONG Pattern:

ORCHESTRATOR → spawns 12 G-2 agents directly → IDE CRASH

The G-2 Commander (G2_RECON) absorbs the parallelism complexity. ORCHESTRATOR only ever spawns 1 coordinator.

---

Spawn Pattern

Via G2_RECON Commander (CORRECT)

# ORCHESTRATOR spawns G2_RECON who manages the 12 G-2 teams
Task(
    subagent_type="general-purpose",
    description="G2_RECON: SEARCH_PARTY Commander",
    prompt="""
## Agent: G2_RECON (G-2 Commander)

You are the G-2 Intelligence Commander for SEARCH_PARTY deployment.

## Mission
Deploy 12 G-2 reconnaissance teams in parallel. Each team runs 10 probes.
Collect all reports and synthesize into unified intel brief.

## Your G-2 Teams to Deploy
1. G2-BACKEND-CORE
2. G2-BACKEND-MODELS
3. G2-SCHEDULING
4. G2-RESILIENCE
5. G2-FRONTEND-CORE
6. G2-FRONTEND-HOOKS
7. G2-TESTS-BACKEND
8. G2-TESTS-FRONTEND
9. G2-MCP
10. G2-DOCS
11. G2-INFRASTRUCTURE
12. G2-SECURITY

## Spawn each using Task tool with subagent_type="Explore"

## After all report back:
1. Cross-reference findings
2. Flag discrepancies (high-signal)
3. Generate consolidated intel brief
4. Report to ORCHESTRATOR
"""
)

Direct Deployment (Only if G2_RECON unavailable)

# Deploy all 12 G-2s in parallel
# WARNING: Only use if spawning from within a coordinator, NOT from ORCHESTRATOR
# Total: 120 probes, wall-clock = single probe timeout

spawn_parallel([
    Task(subagent_type="Explore", description="G2-BACKEND-CORE",
         prompt="Deploy SEARCH_PARTY on backend/app/api/ and backend/app/services/"),
    Task(subagent_type="Explore", description="G2-BACKEND-MODELS",
         prompt="Deploy SEARCH_PARTY on backend/app/models/ and backend/alembic/"),
    Task(subagent_type="Explore", description="G2-SCHEDULING",
         prompt="Deploy SEARCH_PARTY on backend/app/scheduling/"),
    Task(subagent_type="Explore", description="G2-RESILIENCE",
         prompt="Deploy SEARCH_PARTY on backend/app/resilience/"),
    Task(subagent_type="Explore", description="G2-FRONTEND-CORE",
         prompt="Deploy SEARCH_PARTY on frontend/src/components/ and frontend/src/app/"),
    Task(subagent_type="Explore", description="G2-FRONTEND-HOOKS",
         prompt="Deploy SEARCH_PARTY on frontend/src/hooks/ and frontend/src/lib/"),
    Task(subagent_type="Explore", description="G2-TESTS-BACKEND",
         prompt="Deploy SEARCH_PARTY on backend/tests/"),
    Task(subagent_type="Explore", description="G2-TESTS-FRONTEND",
         prompt="Deploy SEARCH_PARTY on frontend/__tests__/"),
    Task(subagent_type="Explore", description="G2-MCP",
         prompt="Deploy SEARCH_PARTY on mcp-server/"),
    Task(subagent_type="Explore", description="G2-DOCS",
         prompt="Deploy SEARCH_PARTY on docs/ and .claude/"),
    Task(subagent_type="Explore", description="G2-INFRASTRUCTURE",
         prompt="Deploy SEARCH_PARTY on docker-compose.yml, .github/, scripts/"),
    Task(subagent_type="Explore", description="G2-SECURITY",
         prompt="Deploy SEARCH_PARTY on backend/app/core/ and backend/app/api/deps.py"),
])

---

Intel Synthesis

After all 12 G-2s report back:

1. Cross-reference findings across domains
2. Flag discrepancies (high-signal findings)
3. Identify gaps requiring follow-up
4. Generate consolidated intel brief

Discrepancy Analysis

Key Insight: Same target, different perspectives. Discrepancies between probes are high-signal:

Discrepancy Type	Signal Meaning
PERCEPTION says X, HISTORY says Y	Recent regression - something changed
INVESTIGATION says connected, INSIGHT says isolated	Undocumented coupling
ARCANA flags violation, PERCEPTION shows green	Silent failure or misconfigured checks
HISTORY shows change, PERCEPTION shows no effect	Change not deployed or cached
INSIGHT expects behavior A, INVESTIGATION shows B	Implementation drift from design

---

Output Format

Per-G-2 Report

## G-2 Intel: [DOMAIN]

### Health Grade: [A-F]

### Key Findings
- [Finding 1 with severity]
- [Finding 2 with severity]

### Discrepancies Detected
- [Probe A vs Probe B disagreement]

### Gaps Identified
- [What wasn't checked]

### Priority Actions
1. [Most urgent fix]
2. [Next priority]

Consolidated Brief

## SEARCH_PARTY Intel Brief (120 Probes Deployed)

| Domain | Health | Issues Found | Priority Findings |
|--------|--------|--------------|-------------------|
| Backend Core | A- | 3 | [findings] |
| Frontend | B | 12 | [findings] |
| Scheduling | A | 0 | Clean |
| ... | ... | ... | ... |

### Cross-Domain Discrepancies
[High-signal findings where domains/probes disagreed]

### Global Gaps
[What wasn't checked across all domains]

### Recommended Actions
1. [Highest priority fix]
2. [Next priority]

---

Timeout Profiles

Profile	Duration	Best For
DASH	60s	Quick triage, P0 emergencies
RECON	120s	Normal reconnaissance (default)
INVESTIGATION	300s	Deep analysis, security audits

---

Failure Recovery

Minimum Viable Intel

Mission can proceed if:

• PERCEPTION (baseline state) ✓
• INVESTIGATION (dependencies) ✓
• ARCANA (domain safety) ✓
• At least 4 of remaining 7 probes

Circuit Breaker

If > 3 consecutive probe failures: Trip to OPEN state, suspend missions.

---

Protocol Reference

Full protocol documentation: .claude/protocols/SEARCH_PARTY.md

---

Related Skills

Skill	When to Use
startup	Session initialization
startupO	ORCHESTRATOR mode initialization
systematic-debugger	Post-recon debugging
parallel-explore	Quick parallel exploration

---

SEARCH_PARTY: Ten lenses, one target, zero marginal cost. The discrepancies are the signal.