What problem does it solve? Executing a fully planned feature across many task beads requires disciplined coordination: claiming work in dependency order, delegating implementation to isolated writers, and independently verifying every change before closing tasks. This Skill automates that entire execution loop so a planned feature is implemented end-to-end without losing state, duplicating writers, or accepting unverified work. ## Core Features & Use Cases - Bead-driven execution loop: Reads ready beads from the Beads manifest, claims them deterministically by dependency and priority, and processes them sequentially through a single reusable Herdr writer target. - Durable execution state: Records base SHA, branch, canonical bead IDs, and per-bead progress in .planning/executions/<feature>/forge-state.json, enabling safe resume after interruption. - Independent verification and evidence gates: Inspects diffs and transcripts, reruns targeted tests, and requires runtime/E2E evidence (numbered screenshots, README mapping) before closing any bead. - Use Case: After completing a Phase 6 plan for a new checkout feature, invoke Forge once and it claims each bead, delegates implementation to a Herdr writer, verifies the results, collects E2E screenshots, and finishes with a full report of changed files, tests, and residual risks. ## Quick Start Ask the assistant to run /onehammer:forge for your planned feature so it implements every bead, verifies each one, and reports the evidence.
