ZERG Select Tool — Details Reference

ZERG Select Tool — Details Reference

Scoring Formulas

Five axes scored 0.0–1.0, combined into a composite for tool selection.

File Count (weight: 0.20)

Analysis Depth (weight: 0.30)

Domain (weight: 0.20)

Parallelism (weight: 0.15)

Interactivity (weight: 0.15)

Composite Formula

composite = (file × 0.20) + (depth × 0.30) + (domain × 0.20) + (parallel × 0.15) + (interactive × 0.15)

Tool Capability Matrix

MCP Servers

Native Tools

Task Agent Subtypes

Tier Mapping

Fallback Chains

When a preferred tool is unavailable, degrade gracefully:

Config Schema

MCP server capabilities in .zerg/config.yaml:

mcp_servers:
  context7:
    enabled: true
    capabilities: [documentation, patterns, version-lookup]
    domains: [frontend, backend, devops]
  sequential:
    enabled: true
    capabilities: [analysis, debugging, architecture]
    domains: [any]
  magic:
    enabled: true
    capabilities: [ui-generation, design-system]
    domains: [frontend, ui]
  morphllm:
    enabled: true
    capabilities: [bulk-edit, pattern-enforcement, style]
    domains: [any]
  serena:
    enabled: true
    capabilities: [symbol-ops, project-memory, lsp]
    domains: [any]
  playwright:
    enabled: true
    capabilities: [browser-test, e2e, visual, accessibility]
    domains: [frontend, testing]

Note: If config uses a simple string list (mcp_servers: [context7, sequential]), the selector applies hardcoded capability defaults matching the table above.

Verbose Output Format

When --verbose is passed, the scoring breakdown is displayed:

╔══════════════════════════════════════════════╗
║          TOOL SELECTION SCORING              ║
╠══════════════════════════════════════════════╣
║ Task: "debug the authentication flow"        ║
╠──────────────┬───────┬────────┬─────────────╣
║ Axis         │ Raw   │ Weight │ Weighted    ║
╠──────────────┼───────┼────────┼─────────────╣
║ File Count   │  0.3  │  0.20  │  0.060      ║
║ Depth        │  0.8  │  0.30  │  0.240      ║
║ Domain       │  0.8  │  0.20  │  0.160      ║
║ Parallelism  │  0.1  │  0.15  │  0.015      ║
║ Interactivity│  0.1  │  0.15  │  0.015      ║
╠──────────────┴───────┴────────┼─────────────╣
║ COMPOSITE                     │  0.490      ║
╠───────────────────────────────┼─────────────╣
║ TIER                          │ Single Tool ║
╠═══════════════════════════════╧═════════════╣
║ Recommendations:                             ║
║  1. Sequential (analysis + debugging)        ║
║  2. Task:Explore (codebase understanding)    ║
║ Fallback: Native reasoning                   ║
╚══════════════════════════════════════════════╝

Worked Examples

Example 1: "debug the authentication flow in the API"

Composite: (0.3×0.2)+(0.8×0.3)+(0.8×0.2)+(0.1×0.15)+(0.1×0.15) = 0.49 Tier: Single Tool → Sequential (debugging + architecture analysis) Fallback: Native reasoning with extended thinking

Example 2: "update all React components to use hooks"

Composite: (0.9×0.2)+(0.5×0.3)+(0.8×0.2)+(0.8×0.15)+(0.1×0.15) = 0.62 Tier: Combination → Morphllm (bulk pattern edit) + Context7 (React hook patterns) Fallback: Individual Edit calls + native React knowledge

Example 3: "test the login form in the browser"

Composite: (0.1×0.2)+(0.5×0.3)+(0.8×0.2)+(0.1×0.15)+(0.8×0.15) = 0.47 Tier: Single Tool → Playwright (browser automation + E2E) Fallback: Unit tests + manual verification

Edge Cases

Task Tracking

All /zerg:select-tool invocations are tracked via the Claude Code Task ecosystem:

TaskCreate  → subject: "[Select] Tool selection: {task_summary}"
TaskUpdate  → status: "in_progress" (scoring phase)
TaskUpdate  → status: "completed" (recommendation delivered)

Results are stored as task metadata for downstream commands (/zerg:rush, /zerg:worker) to reference when configuring MCP servers and agent types.