tool-allocator

Tool Allocator

The tool allocation manager — helps every Agent know which tools to use

---

When to Use This Skill

Activate this skill when you need to:

• Install new tools: "I just installed a new MCP/​Skill, allocate it"
• Check usage: "Which tools are not being used?"
• View allocation: "Show me what tools are assigned to which agents"
• Review inventory: "What tools do I have available?"

---

Quick Start

# Sync all tools (auto-discovers + allocates)
node scripts/​index.js sync

# List current allocation
node scripts/​index.js list

# Check for unused tools
node scripts/​index.js check

---

Commands

---

Example Output

sync command

✅ Tool allocation sync complete

📊 Allocation Overview:
   ┌─────────────────────────┬────────────────────┐
   │ Tool                    │ Assigned To        │
   ├─────────────────────────┼────────────────────┤
   │ playwright              │ AIUIUX, AITA, AICA │
   │ supabase                │ AITA, AICA         │
   │ memory                  │ everyone           │
   └─────────────────────────┴────────────────────┘

🔄 Updated:
   ✓ Agent/​AIBA.md
   ✓ Agent/​AIUIUX.md
   ✓ Agent/​AITA.md
   ✓ Agent/​AICA.md

check command

🔍 Tool Usage Check

✅ In Use:
   - chrome-cdp ← AIUIUX, AITA, AICA
   - playwright ← AIUIUX, AITA, AICA

⚠️ Unused:
   - none found

---

How It Works

1. Discovery → Scan opencode.json + skills directories
2. Analysis → Read SKILL.md, extract capabilities
3. Detection → 4-dim agent role analysis
4. Matching → Match tools to agents
5. Sync → Update config files

---

Prerequisites

• OpenCode environment
• 3+ MCPs or Skills installed
• 2+ agents configured
• Node.js 16+

---

File Structure

tool-allocator/
├── SKILL.md              # This file
├── REFERENCES.md         # Detailed reference
├── tool-allocator.config.yaml
├── scripts/
│   ├── index.js          # Entry point
│   ├── discover.js       # Tool/​Agent discovery
│   ├── analyzer.js       # Capability analysis
│   ├── matcher.js        # Role detection
│   └── allocator.js      # Config sync
├── references/
│   ├── commands.md       # Detailed commands
│   ├── config.md         # Configuration guide
│   └── matching-rules.md # Algorithm deep-dive
├── examples/
│   └── real-usage.md    # Real usage scenarios
└── README.md

---

Configuration

# Custom matching rules
rules:
  "frontend-design": ["frontend", "design"]
  "playwright": ["verification", "frontend"]
  "memory": ["*"]  # * = everyone

# Exclude tools
exclude:
  - "chrome-cdp"

---

Allocation Rules

Two Types of Tools

Matching Examples

⚠️ Don't use * for generic tools - they don't need allocation. Only configure domain-specific rules. |

---

⚠️ Common Gotchas

• Config file must be valid YAML - Use https://www.yamllint.com to validate before editing
• Agent configs must exist before sync - Create agents in opencode.json first
• Memory updates require persistent-memory skill - Skip if skill not installed
• Sync only updates "## Tools" section - Preserves all other content in agent MD files
• Glob/​Search patterns are case-sensitive - Match exact tool names
• Backups use .bak extension - Don't confuse with original files
• Don't allocate by name alone - Decision process: (1) Read SKILL.md description to understand actual domain; (2) Check provider (google-gemini → code, novel-writer → novel); (3) Match to agent's core responsibility. Example: requirement-detector (novel domain) ≠ BA tool, google-gemini-code-reviewer (code domain) = AITA/​AICA.
• Don't force allocation for generic tools - Tools like memory, one-search, cocoloop, persistent-memory, tool-allocator, url-capability-analyzer are "Swiss Army knives" - main agent can use them directly. Only domain-specific tools need allocation.

Edge Cases

• New agent with no MD file → Skip silently, continue with other agents
• Tool already allocated → Skip (no duplicates in allocation)
• Circular permissions → Handled by max depth 3 in permission chain analysis
• Empty opencode.json → Warn but continue with empty discovery
• MCP with no tools listed → Skip MCP, continue with others
• Skill without SKILL.md → Use folder name as fallback identifier
• Duplicate tool names → First discovered takes precedence
• Tools with misleading names → Read description + check provider. "detector" sounds like BA but requirement-detector is for novels; "google-gemini" is code-related even though it could be used for novels.

---

Version History

---

MIT License