What problem does it solve? Building a feature end-to-end requires coordinating decomposition, isolated worktrees, per-task commits, test suites, and independent review, and doing that from memory is error-prone. This Skill drives the whole run through the clerk CLI, which tracks run state and returns the next step with its method on every call. ## Core Features & Use Cases - Clerk-driven step loop: Call clerk step to get the next undone step (ground, isolate, decompose, build, suite, audit, match-request, verify-run, land, learn) with full instructions, so run order is never held in memory. - Direct construction with independent audit: You write the code for every task yourself, then the finished branch goes to audit-implement for adversarial review by readers who never watched it being written. - Resumable runs and repo-level flags: Position is recomputed from the repository and ledger on every call, and flags like --in-place, --integrate, --review-breakdown, and --gears resolve from both the request and repo settings. - Use Case: You receive a feature request whose exact shape is not yet settled. Start a run with clerk step start my-feature --request "..." --harness opencode, follow each returned step to decompose and build the tasks with per-task commits, and let the audit loop review the branch before landing it. ## Quick Start Ask the AI to implement your feature request with the implement skill, passing the request verbatim so clerk can decompose it into tasks and drive the build and audit loop.
