lib-process:meta--github-copilot
assimilation/harness/github-copilot overview
Orchestrate babysitter SDK integration into GitHub Copilot CLI agent. GitHub CLI (`gh`) is GitHub's official CLI tool. It includes: - `gh copilot` subcommand for AI features (suggest, explain) - `gh copilot suggest` / `gh copilot explain` for one-shot AI interactions - `gh extension` system for distributing custom CLI commands as repos - GitHub Copilot in the CLI for agent-style coding (emerging feature) The GitHub CLI is NOT a traditional coding harness like Claude Code or Codex. It's a GitHub API client with Copilot features bolted on. The integration approach differs from other harnesses: - Hook system exists but limited: hooks.json with camelCase types (sessionStart, sessionEnd, userPromptSubmitted, preToolUse, postToolUse, errorOccurred) — but ONLY preToolUse can return flow-control decisions. All other hook outputs are IGNORED. No stop-hook equivalent exists. - Plugin system: manifest in .plugin/, .github/plugin/, .claude-plugin/, or repo root. Install via `copilot plugin install OWNER/REPO`. - Marketplace: marketplace.json, `copilot plugin marketplace add/browse/list` - Plugin storage: ~/.copilot/installed-plugins/ - Skills: skills/NAME/SKILL.md format with frontmatter - Extension model: `gh extension install owner/repo` (binary or script) - Config: ~/.config/gh/ (hosts.yml, config.yml) - Environment: GH_TOKEN, GITHUB_TOKEN, GH_HOST, GH_REPO, etc. - GitHub Copilot Workspace and Copilot Agent may introduce richer surfaces The research phase MUST verify the current state of: - GitHub Copilot agent capabilities (beyond suggest/explain) - Whether `gh copilot` supports agent loops, tool use, or MCP - GitHub Copilot Workspace API/CLI if available - Whether there's a session or conversation model - Any hook or callback mechanism for continuation This is a more speculative integration — the process must determine whether GitHub Copilot CLI has evolved enough to support babysitter orchestration, or if the integration should focus on the extension model as a distribution mechanism with in-turn loop driving.
Attributes
Outgoing edges
- domain:software-engineering·DomainSoftware Engineering
- specialization:meta·SpecializationMeta / Process Engineering
- topic:developer-experience·TopicDeveloper Experience (DX)
- topic:integrations·TopicIntegrations
- skill-area:ai-agent-development·SkillAreaAI Agent Development
- skill-area:orchestration-loop·SkillAreaOrchestration Loop Engineering