← ClaudeAtlas

finish-tasklisted

Finish an implemented and reviewed task — README audit, final behavior proof, completion record, PR creation after approval, and cleanup of the task's retained review files after CI passes. Use manually after the task-review and review-fix-worker loop is clean. Lightweight orchestration suited to a small fast model.
SrdjanCoric/skills · ★ 18 · AI & Automation · score 64
Install: claude install-skill SrdjanCoric/skills
# Finish Task Take one implemented task whose manual review loop is complete and carry it to a CI-green PR. This skill is lightweight orchestration: it updates the README, proves final behavior, records completion, hands off to `create-pr` after explicit user approval, and removes the task's retained review files after the PR is CI-green. It does not implement, review, or remediate. ## Model routing This skill is mechanical orchestration over already-verified work. Run it on the smallest capable fast model (router: `finish-task` maps to the luna-max route). Escalate to the user rather than muscling through ambiguity. ## Preconditions 1. Resolve the task from a supplied path, or find the single `[~]` task in the master plan. With zero or multiple `[~]` tasks and no supplied path, stop and report the state. 2. Confirm the current branch matches the task's `**Branch**` value; otherwise stop. 3. Normalize the selected task to its repository-relative path. Inspect every `reviews/*.md` file at the repository root and select those whose `**Task:**` matches that normalized path. If any matching file contains a finding with `**Status:** open`, stop and tell the user to run `/skill:review-fix-worker` (or explicitly skip or accept the remaining findings) first. Absence of a matching review file for a `code` or `mixed` task means `task-review` never ran; stop and tell the user to run `/skill:task-review` first. Retain the matching file list and each review outco