---

name: search-party description: Parallel 120-probe reconnaissance using G-2 RECON agents. Deploy 12 G-2s commanding 10 D&D-inspired probes each. Use for comprehensive codebase exploration. model_tier: sonnet parallel_hints: can_parallel_with: [systematic-debugger, code-review] must_serialize_with: [] preferred_batch_size: 12 context_hints: max_file_context: 200 compression_level: 2 requires_git_context: true requires_db_context: false escalation_triggers:

• pattern: "security.*breach" reason: "Security findings require immediate escalation"
• pattern: "credentials|secrets" reason: "Credential findings require security review"

---

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.

CopySequential (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:

Probe Composition (Per Party)

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

---

Invocation

Full Deployment (120 probes)

Copy/search-party

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

Targeted Deployment (10 probes)

Copy/search-party backend/app/scheduling/

Deploys single G-2 on specific target.

Quick Search (focused)

Copy/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:

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

WRONG Pattern:

CopyORCHESTRATOR → 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)

Copy# 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)

Copy# 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:

---

Output Format

Per-G-2 Report

Copy## 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

Copy## 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

---

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

---

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