← ClaudeAtlas

deliverlisted

Use when the user hands over one task to be finished end to end with real E2E proof, locally and again on dev after merge — `/deliver <task |
zahidakhyar/skills · ★ 0 · AI & Automation · score 72
Install: claude install-skill zahidakhyar/skills
# Deliver One task, carried to evidence. Phase 1 goes from the task to an open PR with local E2E proof. Phase 2 runs after the user merges and repeats the same scenarios on dev. The merge itself is always the user's. `continue` on a merged PR that never went through phase 1 runs phase 2 alone, with scenarios P1 writes from the PR. **"Done" means every scenario ran and its evidence exists in the run folder.** A scenario that did not run is `UNVERIFIED` in the report, never left out, never described as passing. ## Arguments - `<task text | #N | issue URL>` → phase 1. - `continue [PR]`, or a message saying the PR is merged. Check in this order: 1. No PR given and a blocked run exists (`run_dir.sh --branch <current branch>` with `phase: blocked`, else `run_dir.sh --blocked`, which works from the main checkout) → resume phase 1 at step 4 inside its `worktree`, with its criteria and scenarios. 2. Otherwise → phase 2 for the given PR, else the current branch's PR. Neither exists → ask which PR. - `continue <PR> --with <PR URL> ...` → phase 2 for `<PR>`, together with PRs in other repos that must be live on dev at the same time (`orchestrate` suggests this after a cross-repo build). The URLs go into `state.json` as `with`. ## Run folder `bash ${CLAUDE_SKILL_DIR}/scripts/run_dir.sh <slug>` creates it in the main checkout (`.deliver/`) and excludes it from git. Runs started before this skill was renamed from `autopilot` live under `.autopilot/`; the search options find them t