Implement Plan

Execute a plan artifact end-to-end on the current working tree. This skill reads the plan artifact READ-ONLY, reads the plan index, walks its task files in plan order, implements each task, self-reviews after each task, auto-commits per plan task, records a factual progress block per plan task in a run progress file, and updates the thread's singleton implementation report on the way out. It does not pause for clarifying questions at each step and does not ask before committing; the execution posture is identical whether or not a person is present. It does not rewrite history.

Inputs

This skill accepts a plan artifact. The plan is a multi-file artifact rooted in the active thread folder:

docs/threads/<YYMMDDHHMMSSZ-slug>/
├── plan.md                 # the index, at the thread root
└── plan-tasks/
    ├── 01-<kebab-slug>.md
    ├── 02-<kebab-slug>.md
    └── …

The index plus its plan-tasks/ folder together are the plan artifact. plan.md is the index: it carries the plan-level objective and context, a Source: line naming the upstream artifact the plan was compiled from (a thread-relative pointer, a repo-relative path, an issue URL, or none — raw prompt), a Global Constraints block, and an ordered task list that is authoritative for task count and order. Each plan-tasks/NN-<kebab-slug>.md file is one task, carrying the six mandatory fields — Objective; Input / context; Steps / substeps; Files modified; Verification; Acceptance criteria — plus two hand-off lines, Consumes: (what this task uses from earlier tasks) and Produces: (what later tasks rely on), where none is a legal value for either. Every task must be sequential, isolated, independently implementable, and independently reviewable, and the tasks are executed in index order.