Research: Copilot CLI native integration strategies

[abeltrano]

Summary

Research how to integrate PromptKit natively into GitHub Copilot CLI so it feels like a built-in capability rather than a separate addon. The current workflow requires users to "read and execute bootstrap.md" or run a separate npx promptkit CLI that spawns a child Copilot process — both feel disjointed.

Pain Points Identified

• Requires separate CLI installation
• Spawns a child Copilot process, losing existing session context
• No integration with Copilot's discoverability (/skills, /agent, etc.)
• Cannot be invoked naturally within an existing session
• Context discontinuity: When a user generates a PromptKit artifact (e.g., a custom agent) and later needs another PromptKit capability (e.g., bug investigation), they must break out of their current session, losing accumulated context

Research Scope

Evaluate all Copilot CLI extension points for PromptKit integration:

1. Skills — on-demand task instructions (/promptkit)
2. Custom Agents — specialized personas with infer: true auto-delegation
3. MCP Server — deterministic, programmatic assembly
4. Plugins — distribution vehicle bundling all components
5. Hooks — lifecycle guardrails and output validation
6. LSP Configs — enhanced code intelligence for code-analysis templates
7. Custom Instructions — lightweight always-loaded context

Deliverables

• Research document (docs/copilot-cli-integration-research.md)
• Roadmap update (docs/roadmap.md)
• Requirements document
• Design document
• Validation plan
• Implementation plan