What problem does it solve? When an AI-assisted coding session ends mid-task, the next session starts from zero: it does not know which phase the task reached, which branch or worktree holds the code, or whether the previous session left the repo in a working state. This Skill reconstructs that context from the task folder so work continues without redoing or forking prior progress. ## Core Features & Use Cases - Phase Detection: Reads task artifacts (raw_prompt.md, prompt-understanding.md, execution_plan.md, acceptance_criteria.json) to determine which of phases 0-4 the task is in and what the next action is. - Baseline Smoke Test: Boots the project and re-runs touched-path tests plus previously-passing acceptance criteria before any new code is written, surfacing regressions left by the prior session. - Worktree-Aware Branch Recovery: Resolves the remote default branch and reconciles recorded worktree fields against the current HEAD, preventing accidental work on main or forking into a duplicate branch. - Use Case: You return Monday morning to a half-finished feature. You give the assistant the task folder path, it reports the current phase, confirms the baseline is green, and continues coding from the next failing acceptance criterion. ## Quick Start Say "resume task" and provide the path to your task folder, for example tasks/my-namespace-195-my-task, then confirm the resumption summary.
