What problem does it solve?
Execute turns a plan (or intent-derived requirements) into an end-to-end, guardrailed build workflow by orchestrating task execution, verifying outcomes, applying contract auto-patches, and producing a structured stdout report—so you don’t have to manually coordinate steps or trust unverified outputs.
Core Features & Use Cases
- Plan-first orchestration: Normalizes inputs, validates/initializes a plan.json, and dispatches worker work according to plan structure and dependencies.
- Configurable dispatch + verification depth: Prompts you for dispatch mode (Direct/Agent/Team) and verify depth (Light/Standard/Thorough) and applies gate-based verification accordingly.
- Deterministic orchestration with resumability: Treats plan.json as the ledger, skips tasks already done, enforces ceilings, and supports idempotent restart behavior.
- Contracts mismatch auto-patching: When worker outputs indicate contract/interface drift, auto-edits contracts.md (without user confirmation) and logs the change to audit.md.
- Coverage-driven verification and reporting: Runs unified verification, performs gap re-dispatch when citations are missing, and prints an ordered 8-section report to stdout.
Quick Start
Ask the AI to run execute with your plan directory by saying: run execute in the spec folder that contains plan.json or requirements.md, and choose Direct dispatch and Standard verify depth.