← ClaudeAtlas

implement-tasklisted

Use when implementing a software task end-to-end — triggers include "implement this", "build this feature", "implement this ticket", a Jira ID/URL, or a plan file path. ALWAYS use for end-to-end implementation work.
atomgunlk/skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill atomgunlk/skills
# implement-task End-to-end workflow for implementing a software task — from a task source (Jira / plan doc / prompt) to an open PR. Brainstorms with the user, has the approved plan implemented test-first in a git worktree (delegated to implementer subagents by default — the controller reviews, it doesn't write), then runs a whole-branch Review-Fix loop that repeats until no Critical/Important findings remain. This skill is **project-agnostic**: it carries the _method_, the repo carries the _facts_. In each repo, discover conventions and commands from the project itself rather than assuming them. This skill is **procedural and ordered**. Do not skip or reorder phases. Each phase has a checkpoint. If a change spans multiple repos, run the per-repo phases (2, 3, 4, 5, 6, 7) **once per repo** — the approved plan (Phase 1) decides which repos those are. **REQUIRED SUB-SKILLS** (invoke them via the `Skill` tool — don't reinvent): - `superpowers:brainstorming` — collaborative planning with the user before any code is written (Phase 1). - `use-worktree` — create the branch + sibling worktree via its bundled `scripts/create-worktree.sh` (Phase 2). - `superpowers:test-driven-development` — implement the plan test-first; production code and its tests ship in the same commit (Phase 3). - `superpowers:verification-before-completion` — evidence before any "done" / "passing" claim (Phase 5). - `pr-review-toolkit:review-pr` — the controller's single whole-branch review-fix loop until n