exe-plan

Dispatch plan phases to subagents using plan.md as the single source of truth.

Updated Aug 27, 2026
One-click install
npx skills add https://github.com/gsa9/claude-plan-skills --skill exe-plan
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: exe-plan
Source: https://github.com/gsa9/claude-plan-skills/tree/main/skills/exe-plan
Command: npx skills add https://github.com/gsa9/claude-plan-skills --skill exe-plan

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

This Skill enables executing a multi-phase plan by delegating each phase to independent subagents, while the main thread acts as a minimal orchestrator.

Core Features & Use Cases

  • Lightweight orchestration: Reads the plan.md once and dispatches phases to subagents without executing tasks itself.
  • Deterministic execution: Supports parallel groups and sequential mode (--sequential) to control phase concurrency.
  • Resumable flow: Tracks completion via plan-results/done-phase-*.md to resume after failures.

Quick Start

Use the /exe-plan command to run the current plan.md in a fresh session, ensuring plan integrity and isolated subagent execution.

Frequently Asked Questions about exe-plan

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

FAQPage Schema
How do I execute a multi-phase plan using subagents in a fresh session?

To execute a multi-phase plan using subagents, the main thread reads plan.md once and dispatches phases to subagents via the Task tool while tracking completion with done-phase-*.md files. This lightweight orchestration approach isolates execution context and ensures deterministic phase processing without polluting the main session.

What is the best way to orchestrate parallel task execution for a markdown plan file?

Orchestrating parallel task execution for a markdown plan file involves dispatching grouped phases to independent subagents simultaneously. The system uses plan.md as the single source of truth and supports parallel grouping natively, with an optional --sequential flag to force strict phase-by-phase execution when needed.

Can I resume a workflow after a subagent task dispatch fails midway through plan execution?

Yes, you can resume a workflow after a subagent task dispatch fails because the orchestrator tracks completion via plan-results/done-phase-*.md files. These markers allow the system to identify finished phases and resume execution from the exact point of failure in a subsequent run.

Does the orchestrator execute plan steps directly or only delegate them to subagents?

The orchestrator only delegates plan steps to subagents and does not execute tasks directly. It acts as a lightweight coordinator that reads plan.md, manages phase dispatch using the Task tool, and monitors progress, ensuring the main thread remains isolated from the actual execution logic.

When should I use the sequential flag instead of parallel grouping for task dispatch?

Use the --sequential flag instead of parallel grouping when plan phases have strict dependencies that require each step to complete before the next begins. Parallel grouping is better suited for independent phases that can run concurrently, maximizing throughput in the subagent execution workflow.

Why does the plan execution workflow require a fresh session without brainstorm context?

Plan execution requires a fresh session without brainstorm context to maintain plan integrity and ensure deterministic phase execution. Isolating the workflow from previous brainstorming or create-plan context prevents interference and allows subagents to operate cleanly on the plan.md source of truth.