What problem does it solve? Turning a written plan or a bare prompt into shipped, tested code is error-prone: context degrades across many tasks, tests get skipped, and verification is inconsistent. This Skill executes work documents systematically with task triage, subagent dispatch, incremental commits, and layered self-verification so features are completed with quality gates enforced. ## Core Features & Use Cases - Plan-Driven Execution: Reads a plan, spec, or todo file (or a bare prompt), triages complexity, builds a task list, and executes inline or via serial/parallel subagents. - Agent Teams Mode ([T]/[T+]): Spawns a verifier teammate (and a risk-guard in [T+]) that independently validates each unit via SendMessage before work proceeds. - Four-Layer Self-Verification ([V]/[V+]): Auto-detects the project's start and test commands (environment fingerprint persisted in CLAUDE.md), runs tests, and optionally uses Playwright MCP for browser-level validation. - Use Case: After running /ce:plan for a new feature, invoke /ce:work with the plan path and [V] to implement each unit, run tests continuously, commit incrementally, and finish with an embedded ce:review quality pass. ## Quick Start Ask the AI to execute the plan at docs/plans/my-feature.md using ce:work with the [V] flag so it implements each unit, runs tests, and verifies the result automatically.
