← ClaudeAtlas

task-implementlisted

Execute a ready spec interactively in the current session — implements the spec following its Chosen Approach (feature) or fix approach with regression test (bugfix), then auto-chains into /marvin:task-verify and /marvin:task-deliver. Use when the user says "run the task", "execute the spec", "implement this spec", "/marvin:task-implement", "marvin implement the spec", or after /marvin:task-start produced a spec and they want to implement it without dispatching to a worktree.
real-case/marvin-toolkit · ★ 0 · AI & Automation · score 68
Install: claude install-skill real-case/marvin-toolkit
# Run Execute a spec that passed the Definition-of-Ready gate. Runs interactively in the current session on the current branch — you see every step and can intervene. On success, the skill auto-chains into `/marvin:task-verify` and `/marvin:task-deliver`. ## Core principle **The spec is the instruction set.** This skill is the interactive sibling of `marvin-tm-executor` (which runs the same spec headless, dispatched via Task-tool). Same contract, same pipelines — but the human is in the loop and the result stays on the current branch until `/marvin:task-deliver` opens the PR. ## Input `$ARGUMENTS` — one of: - A spec file path (`.marvin/task/<NNN>-<slug>.md`) - A slug (`<slug>` → resolves to the spec whose filename is `<slug>.md` or `<NNN>-<slug>.md`) - Empty — the skill resolves the spec per Step 1 --- ## Workflow ### 1. Resolve the spec **Where specs live.** Call the `spec` MCP tool with `action: "list"`. It answers with the resolved spec directory — configured, detected, or the `.marvin/task/` default — and every spec in it, each with its slug, title, status and seal. That answer is "the spec corpus" below; do not enumerate directories by hand. A slug resolves to a file prefix-tolerantly (`<slug>.md`, else `<NNN>-<slug>.md`), which is the contract, not something to re-implement. Resolution order: 1. **Argument provided:** - If it ends in `.md` and the file exists — use it. - Else treat as a slug and take the matching record from the corpus. If there is none