integration-discovery

Capability Router

Purpose: Scout available capabilities and decide what to use for a task. Return actionable recommendations so the main agent can execute without discovery overhead.

Why This Matters

Most agents execute with whatever tools are obvious. They don't:

• Check what else is available
• Consider skills that could help
• Look for specialized agents
• Adapt to the user's configured integrations

This skill changes that. It scouts, evaluates, and recommends—so the main agent can focus on execution.

Quick Usage

As Inline Check

Before executing this task, let me check what capabilities could help.

**Task**: Review PR #123 for security issues

**Discovery**:
- MCP: github (PR context), atlassian (link to Jira)
- Skills: qe (test scenarios), platform (security checks)
- Agents: wicked-garden:platform:security-engineer

**Recommendation**: Use security-engineer agent for review, github MCP for PR data.

As Subagent

For complex tasks, spawn integration-discovery as a subagent:

Task tool:
  subagent_type: "Explore"
  prompt: "Scout available capabilities for: [task description].
           Return specific recommendations on which MCP servers,
           skills, and agents to use."

The Discovery Process

Step 1: Understand the Task

What does this task need?

• Information sources: Where does data come from?
• Actions: What operations are required?
• Quality concerns: Security, testing, compliance?
• Output: Where do results go?

Step 2: Scout Available Capabilities

Check these sources (see refs/discovery-sources.md for details):

Source	How to Check	What You Get
MCP Servers	ListMcpResourcesTool	External integrations
CLI Tools	command -v {tool}	Installed binaries in PATH
Skills	Check Skill tool description	Methodology/expertise
Agents	Check Task tool agent list	Specialized workers
Built-in Tools	Known set	File ops, search, web

Step 3: Match Capabilities to Needs

For each task need, find matching capabilities:

Task: "Implement user authentication"

| Need | Capability | Recommendation |
|------|------------|----------------|
| Security review | wicked-garden:platform:security-engineer | Use for auth review |
| Test strategy | wicked-garden:crew:gate-adjudicator | Use for test planning |
| Store decisions | wicked-brain:memory | Store auth decisions |
| Track work | MCP:atlassian or native TaskCreate | Track implementation |

Step 4: Return Recommendations

Output a clear recommendation the main agent can act on:

## Capability Recommendations for: [Task]

**Use these**:
1. **wicked-garden:platform:security-engineer** - Review for security issues
2. **MCP:atlassian** - Link work to Jira ticket AUTH-123

**Available but optional**:
- wicked-garden:crew:gate-adjudicator - If test planning needed
- wicked-brain:memory - Store decisions for future reference

**Not available** (consider installing):
- No analytics MCP found - can't check auth failure metrics

Common Task Patterns

Quick mappings for common tasks (see refs/task-patterns.md for full list):

Task Type	Likely Capabilities
Code review	engineering agents, github MCP
Security audit	platform agents, qe
Architecture design	engineering agents, wicked-brain:memory
Bug investigation	engineering agents, error tracking MCP
Feature planning	product agents, project management MCP
Documentation	engineering agents, confluence MCP

Decision Principles

When multiple options exist:

1. Prefer specialized over general - Use security-engineer for security, not generic reviewer
2. Prefer available integrations - Use configured MCP over manual alternatives
3. Consider task scope - Simple task? Skip the heavy machinery
4. Check dependencies - Some agents work better together

Output Format

Always return structured recommendations:

## Capability Recommendations

**Task**: [one-line task description]

### Recommended
| Capability | Type | Use For |
|------------|------|---------|
| name | MCP/Skill/Agent | specific purpose |

### Optional
| Capability | Type | Use If |
|------------|------|--------|
| name | type | condition |

### Not Available
| Need | Suggestion |
|------|------------|
| what's missing | how to get it |

References

• Discovery Sources - Where to look and how
• Task Patterns - Common task-to-capability mappings
• MCP Capabilities - MCP server categories and patterns
• CLI Detection - CLI tool detection patterns and decision policy