Agent Discovery Skill

The Agent Discovery skill enables structured scanning of agent repositories, marketplaces, and research platforms to discover new AI agents for potential integration into G-Pilot.

Overview

This skill follows the Action-Ledger-Result pattern and integrates with:

• GitHub API (repository scanning)
• HuggingFace API (model discovery)
• AI-powered analysis for other sources (LangChain Hub, CrewAI, etc.)

Bound Functions

agent_discovery

Primary discovery function - Scans multiple intelligence sources for agents.

agent_discovery(userId, query, options) → DiscoveryResult

Parameter	Type	Description
userId	string	User ID for billing and auditing
query	string	Search query
options.sources	string[]	Sources to scan (defaults to top 3)
options.limit	number	Max results per source
options.focusAreas	string[]	Capability areas to prioritize
options.minCompatibility	number	Minimum compatibility score (0-1)

check_compatibility

Compatibility assessment - Deep analysis of a single agent dossier.

check_compatibility(userId, dossier) → CompatibilityCheckResult

Intelligence Sources

Source	API Type	Capabilities
github	REST API	Repository metadata, stars, license
huggingface	REST API	Model info, downloads, pipeline tags
langchain_hub	AI-scanned	Prompts, chains, agent templates
crewai_flows	AI-scanned	Multi-agent workflows
arxiv	AI-scanned	Research papers, techniques
gpt_store	AI-scanned	Custom GPT agents
vertex_ai	AI-scanned	Enterprise agent solutions
aws_bedrock	AI-scanned	Foundation model agents
azure_ai	AI-scanned	Cognitive service agents

Output Schemas

AgentDossier

{
  "agentName": "example-agent",
  "source": "github",
  "sourceUrl": "https://github.com/example/agent",
  "description": "What this agent does",
  "compatibilityScore": 0.85,
  "securityRating": "A",
  "uniqueCapabilities": ["capability1", "capability2"],
  "requiredDependencies": ["node", "typescript"],
  "integrationEstimate": "1-2 days",
  "fuelCostEstimate": "50-100 PTS",
  "lastUpdated": "2026-01-19T00:00:00Z",
  "stars": 1234,
  "maintainer": "example-org",
  "license": "MIT"
}

DiscoveryResult

{
  "success": true,
  "query": "marketing automation agent",
  "sourcesScanned": ["github", "huggingface", "langchain_hub"],
  "dossiers": [...],
  "duration": 2500
}

CompatibilityCheckResult

{
  "success": true,
  "agent": "example-agent",
  "score": 0.8,
  "checks": [
    { "name": "Language Compatibility", "passed": true },
    { "name": "License Check", "passed": true },
    { "name": "Security Rating", "passed": true }
  ],
  "recommendation": "integrate"
}

Usage Example

import { agent_discovery, check_compatibility } from '@/tools/agentDiscoverySkill';

// Discover agents for a specific capability
const result = await agent_discovery(
  'user_123',
  'multi-agent orchestration',
  {
    sources: ['github', 'langchain_hub'],
    limit: 10,
    focusAreas: ['automation', 'workflow'],
    minCompatibility: 0.7
  }
);

// Deep check a specific candidate
for (const dossier of result.dossiers.slice(0, 3)) {
  const check = await check_compatibility('user_123', dossier);
  if (check.recommendation === 'integrate') {
    console.log(`Recommended: ${dossier.agentName}`);
  }
}

Integration with Agent Scout

This skill is the primary tool bound to the Agent Scout agent:

Agent Scout
    ├── agent_discovery (primary)
    ├── check_compatibility
    ├── web_intel
    ├── structured_scraper
    └── deep_lookup

Security Ratings

Rating	Description	Criteria
A	Trusted	Popular, well-maintained, permissive license
B	Safe	Known maintainer, permissive license
C	Review	Limited info, needs manual review
D	Caution	Unknown maintainer, restrictive license
F	Reject	Security concerns, deprecated

Billing

Operation	Fuel Cost
Quick Scan (3 sources)	25 PTS
Deep Scan (all sources)	100 PTS
Individual Compatibility Check	10 PTS