execute-plan

Executes approved plans phase-by-phase using worktrees, code validation, and feedback triage cycles.

Updated Mar 21, 2026
One-click install
npx skills add https://github.com/mikestopcontinues/agent-skills --skill execute-plan-mikestopcontinues
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: execute-plan
Source: https://github.com/mikestopcontinues/agent-skills/tree/main/claude/skills/execute-plan
Command: npx skills add https://github.com/mikestopcontinues/agent-skills --skill execute-plan-mikestopcontinues

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Executing a multi-phase implementation plan by hand means juggling worktrees, code reviews, feedback triage, and merge discipline manually, with no durable record of progress if work is interrupted. This Skill orchestrates the entire per-phase execution lifecycle so an approved plan lands incrementally with validated, reviewable diffs. ## Core Features & Use Cases - Sequential phase orchestration: Parses the plan's tasks.md, resumes from the first unchecked phase, and never starts phase N+1 before phase N merges. - Worktree-managed code dispatch: Creates a named git worktree per phase, dispatches /write-code workers into it, and validates the phase diff with /validate-code using contextual reviewers (security, performance, integration). - Convergence loop: Drives the triage-feedback / process-feedback / lock-decisions cycle until a validation pass returns no actionable clusters, then fast-forward merges and checks off tasks.md. - Use Case: A captain says "execute p018" — the Skill reads the plan, lands each phase in its own worktree with validated diffs, updates tasks.md after every merge, and stops with a summary of landed SHAs and deferred items. ## Quick Start Ask the agent to execute plan p018 and it will walk each phase through worktree setup, code dispatch, validation, and merge until the plan's tasks are all checked off.

Frequently Asked Questions about execute-plan

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

FAQPage Schema
How do I execute a multi-phase implementation plan with an AI agent?▼

Invoke the execute-plan skill with a plan ID like pNNN. It parses the plan's tasks.md, dispatches code-writing workers into a git worktree per phase, validates each phase diff, and merges sequentially until all task rows are checked off.

How does execute-plan resume after an interrupted run?▼

State persists in the plan's tasks.md checked-row markers plus the filesystem. On restart, the skill skips phases whose rows are all checked and picks up any existing worktree for the resume-point phase instead of recreating it.

Can execute-plan run multiple phases of a plan in parallel?▼

No. Phases are strictly sequential — phase N must merge before phase N+1 starts, per the plan's declared ordering. Cross-plan parallelism is possible only by running separate execute-plan invocations on different plans, managed by the captain.

What happens when phase validation finds problems in the code?▼

The skill runs a triage-feedback cycle: clusters routed to act trigger code fixes or plan revisions, process clusters dispatch parallel feedback processors, and decision-shaped clusters go to lock-decisions. Validation repeats until a pass returns no actionable clusters, with a minimum of two iterations.

Why does the execute-plan merge fail with a fast-forward error?▼

A --ff-only failure means the rebase did not catch the current base tip, usually because another landed phase advanced the base. The skill stops and surfaces the conflict to the captain rather than auto-resolving or falling back to a merge commit.

When should I use write-code directly instead of execute-plan?▼

Use write-code directly for ad-hoc code work that is not scoped to a phase within an approved plan. execute-plan is only for executing plans stored under docs/plans with a task breakdown; it refuses plans still in proposal or open-questions posture.