core¶
The core pack ships the agent-ready-repo skills (work-loop, new-spec, new-rfc, …) — including frontend-engineering, the depth skill the work-loop loads inline whenever a task's primary output is HTML, CSS, or JS — the four specialist subagents (adversarial-reviewer, security-reviewer, quality-engineer, implementer), the canonical session-start and work-loop-check hooks, and the /adapt-to-project command.
Install routes and adaptation¶
core ships through three install routes — agentbundle install (CLI), claude plugin install (Claude-plugins), and apm install (APM). Whichever route an adopter uses, the install→adapt chain ends the same way: a [[packs-installed]] entry lands in the scope-correct .adapt-install-marker.toml, the next session of a HookIntegrator-covered target tool surfaces a "N pack(s) pending adaptation; run /adapt-to-project" nudge, and the skill adapts the brownfield repo (or user-home for user-scope packs) to the project.
APM's HookIntegrator projects the SessionStart install-marker hook to four of seven supported target tools: Claude Code, Copilot, Cursor, and Gemini. The remaining three — Codex, OpenCode, and Windsurf — silently lack the hook surface in upstream APM today; their adopters run the documented manual fallback once after install:
This is the same gesture CLI-route adopters use, and the same gesture adopters at HookIntegrator-covered targets can run if they prefer to opt out of hooks for safety. See docs/specs/apm-install-route-parity/spec.md for the contract surface and docs/rfc/0010-apm-install-route-parity.md for the design rationale.
Usage¶
core is the loop itself. Ask your agent, for example:
- "What should I work on next?" (
workspace-status— orient at session start) - "Start a new spec for a rate-limiting feature." (
new-spec) - "Implement this spec with the work-loop." (
work-loop) - "Fix this bug: the importer drops the last row." (
bug-fix) - "Adapt this repo to the installed packs." (
/adapt-to-project)
→ Go deeper: the core guides.