What problem does it solve? Long implementation sessions burn context on file reads, failed edits, and test output, and plan state drifts from reality when steps are marked done without evidence. This Skill drives an existing plans/active/<slug>.yaml to completion while keeping the main session thin: subagents absorb the heavy work and return only verified results. ## Core Features & Use Cases - Frontier computation and clustering: Reads plan state via evo harness show and evo harness step --no-input, computes the ready frontier from depends_on, respects blocking steps and checkpoints, and groups steps into parallel-safe clusters with explicit serial justifications. - Subagent contract with verbatim evidence: Each subagent receives the step's what, verify, and note verbatim, is forbidden from editing plan YAML or self-marking done, and must return the exact verify command with the last 15 lines of unmodified output. - Evidence-driven state writes: The main session validates reports, re-runs only missing or invalidated checks, commits staged changes after secret scan and gitnexus detect-changes, then records status and evidence with evo harness step (never --note, which would overwrite existing notes). - Use Case: You have a 20-step migration plan under plans/active. Invoke the skill with the slug; it runs independent steps in parallel subagents, verifies each, commits, updates the YAML, runs evo harness check, and offers a /clear between clusters with the exact re-entry command. ## Quick Start Ask the agent to execute the plan for a given slug, for example: run the execute-plan skill on plans/active/deployments-control-plane.yaml and continue until all steps are done.
