DevAgentRadar Skill

Use this skill whenever asked to fetch DevAgentRadar action packs, check a project's AI coding agent/harness stack against live DevAgentRadar proposals, or prepare harness upgrade rules:

Workflow Instructions

1. Inspect Workspace Agent/Harness Footprint:

   • Read the project's root package.json (or pnpm-workspace.yaml) for agent CLI packages (e.g. @anthropic-ai/claude-code, aider).
   • Read agent configuration files (.claude/, .cursor/, AGENTS.md, CLAUDE.md) to detect which harnesses and config surfaces (hooks, MCP, permissions, skills) are in use.

2. Query DevAgentRadar MCP Server:

   • Call scan_workspace_health with the extracted agent/harness dependencies and detected surfaces. DevAgentRadar's catalog is AI coding agents & harnesses, not general npm frameworks — anything else comes back as untrackedDependencies rather than a false "all clear".
   • Call get_action_pack for any active proposal slugs to retrieve paste-ready harness rules.
   • Call get_catchup_action_pack to get version catch-up diffs for the tracked stack.
   • Query get_model_radar or search_radar_insights for model releases and architectural guides.

3. Output Action Packs & Guidance:

   • Present 🚨 Breaking Alerts: Critical ecosystem changes affecting this repo's tracked agent/harness stack.
   • Present 📦 Paste-Ready Action Packs: Output harness configuration diffs for .cursor/mcp.json, .claude/settings.json, or AGENTS.md rules.
   • Note any untrackedDependencies so the developer knows what DevAgentRadar didn't (and doesn't) cover.
   • Conclude with a clear Upgrade Verdict (SAFE TO UPDATE, PROCEED WITH CAUTION, or HOLD UPDATE — BREAKING CHANGES DETECTED).