task-execute

Executes implementation plans via git worktrees, subagents, reviews, and PR creation.

3|1|Updated Dec 9, 2025
One-click install
npx skills add https://github.com/isizono/calm --skill task-execute-isizono
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: task-execute
Source: https://github.com/isizono/calm/tree/main/.claude/skills/task-execute
Command: npx skills add https://github.com/isizono/calm --skill task-execute-isizono

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Turning a written implementation plan into reviewed, merged code requires many coordinated steps—branch setup, implementation, code review, commit, and PR creation—that are error-prone when done ad hoc. This Skill automates that execution flow from an existing plan.md through to a pull request. ## Core Features & Use Cases - Autonomous subplan selection: Reads the PR-split table in plan.md, picks the next unstarted subplan whose dependencies are complete, and marks it in progress. - Worktree-based implementation: Creates a git worktree under .trees/, then delegates coding to a background implementer subagent (sonnet) and review to a reviewer subagent (opus) with structured severity findings. - Gate-checked reporting and PRs: Generates a GO decision package, runs size checks before PR creation, reports to the user for approval, and updates plan.md status. - Use Case: After planning a multi-PR feature with task-plan, ask the AI to execute the plan; it selects the next subplan, implements and reviews it in an isolated worktree, reports the results, and opens a PR once you approve. ## Quick Start Execute the next unstarted subplan in plan.md and report back before creating the PR.

Frequently Asked Questions about task-execute

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I execute an implementation plan with AI subagents?▼

Provide a plan.md containing a PR-split table with dependency tracking. The Skill selects the next unstarted subplan whose dependencies are complete, creates a git worktree, delegates implementation to a background subagent, then runs a separate review subagent before reporting results.

What happens if plan.md does not exist when running task-execute?▼

The Skill does not proceed without plan.md. It instructs you to run the task-plan skill first, since planning and execution are separate responsibilities and task-execute only covers worktree creation through PR creation.

How does the automated code review decide pass or rework?▼

The reviewer subagent outputs findings with severity levels (Critical, Major, Minor) but no final verdict. The main agent decides: zero Critical and Major means PASS, small fixes become DIRECT_FIX, and two or more Critical findings trigger re-delegation to a fix subagent.

Can task-execute create a PR without user approval?▼

No. The Skill commits changes and generates a GO decision package, then reports to the user with a structured template. Only after explicit user OK does it run the PR size check, push the branch, and create the pull request.

Why does task-execute use git worktrees instead of branches?▼

Worktrees isolate each subplan's implementation in a separate working directory under .trees/ based on the latest origin/main or a specified base branch. This prevents interference with the main working tree and supports stacked subplans on dependency branches.

What are the limitations of the pull_precedents step?▼

The precedent-pull step depends on an MCP tool that may not be running in all environments. If unavailable, the step is skipped and the GO package's pull.presented field remains 'unavailable', which biases decisions toward pre-GO human review.