What problem does it solve? Building features with an AI agent often loses track of where work stands between sessions, leading to half-finished slices, stale progress files, and unreviewed specs. This Skill turns the feature folder itself into the source of truth, so every session resumes exactly where the work actually is. ## Core Features & Use Cases - Stage routing via a deterministic oracle: Runs speccle next to derive the current stage (plan, spec, implement, stale-claims, done) from the feature folder without calling an LLM. - One human gate: Plan and spec share a single session with one approval point, after which implementation runs unattended and commits on a green checks-gate. - Sibling skill orchestration: Delegates to plan-feature, spec-feature, and implement-feature, and hands off to carve-feature or review at loop boundaries. - Use Case: You hand over a ticket describing a new checkout behavior. The Skill plans and specs it with you in one session, then subsequent sessions implement one criterion at a time, each committed on green, until the slice is done. ## Quick Start Ask the agent to speccle this feature description into a planned, specced slice and start implementing it one criterion at a time.
