Skills
Review patterns an AI agent follows when changing Action
These are project-specific skills or task patterns an AI agent should follow.
Native runtime review
Use when:
- changing AppKit lifecycle code
- changing agent startup or bridge logic
- changing WebKit entry paths
Instructions:
- read Native runtime
- verify whether the change belongs in app code or agent code
- keep AppKit-owned behavior in
Action.app - verify the signed app still builds
Recording review
Use when:
- changing
ScreenCaptureKitrecording - changing stop-file or finished-file handling
- changing probe launch behavior
Instructions:
- read Recording
- inspect the probe launcher and probe app runner
- preserve
.mov+.finishedbehavior - test a real recording path, not just a unit-level refactor
Agent API review
Use when:
- adding WebSocket methods
- changing request or response shape
- updating CLI-to-agent behavior
Instructions:
- read Api
- update the protocol source of truth
- confirm method names remain documented
- verify callers still match the transport contract
Hermes Action
Use when:
- connecting Hermes to the native Action runtime
- using Action MCP tools from a harness
- recording or inspecting native macOS surfaces through Hermes
Instructions:
- read docs/harnesses/hermes-action.md
- use skills/hermes-action/SKILL.md as the harness doctrine
- prefer observe and resolve before act
- treat recording start as asynchronous until
.finishedoraction.record.statusconfirms completion