Guide · Agent skill
How to use the Do Plan Orchestrator agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Do Plan Orchestrator is an agent skill, a packaged set of instructions that tells an AI coding agent how to behave for a certain kind of task. This one, named "do" in the source (from thedotmack/claude-mem, plugin/skills/do, Apache-2.0 licence), makes the agent coordinate the execution of a multi-phase implementation plan instead of making every change itself. This guide explains what the skill does, how to install it, how to trigger it, and what to watch for before you let it commit and push code.
Do Plan Orchestratorthedotmack/claude-mem · Apache-2.0 Download the skill Quick start
- 1Download the ZIP. Get the skill ZIP from the page where you found this guide.
- 2Unzip into a skills folder. Place the skill folder containing SKILL.md in ~/.claude/skills/ for personal use, or in .claude/skills/ inside a project.
- 3Prepare a working branch. Check out a working branch in a Git repository where you have push access, not a production or shared branch.
- 4Make or pick a phased plan. Use a plan with clear phases, ideally one created by make-plan.
- 5Ask in plain language. Request that the agent execute, run or carry out the plan using the do skill.
- 6Review each verified phase. Check the subagent evidence and the commits and pushes after each phase passes verification.
What is an agent skill, and what does the do skill make the agent do?
An agent skill is a folder containing a SKILL.md file with instructions and a short description. The agent reads the description and loads the instructions when a request matches it. The skill itself is a plain text file, so you can read exactly what it asks the agent to do.
The do skill turns the agent into an orchestrator. Instead of editing files directly, the agent deploys subagents, which are separate helper agents, to carry out the work. For each phase of a plan, the source describes an Implementation subagent, then a Verification subagent to prove the phase worked, an Anti-pattern subagent that searches for known bad patterns from the plan, and a Code Quality subagent that reviews the changes. A Commit subagent runs only after verification passes, and a Branch/Sync subagent pushes to a working branch after each verified phase.
- Implementation subagents must copy patterns from documentation rather than invent them.
- Subagents must report evidence such as commands run, outputs and files changed.
- If an API seems to be missing, the subagent is told to stop and verify rather than assume it exists.
- The orchestrator does not advance to the next step until the subagent reports completion and the orchestrator confirms it matches the plan.
How do you install the do skill?
Download the skill ZIP from the page where you found this guide. Unzip it so the skill folder (containing SKILL.md) sits in one of these locations. For a personal install for all your projects, use ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder.
If you use the Claude apps instead of Claude Code, you can upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/ (applies across projects).
- Project install: .claude/skills/ inside the project folder (applies to that project only).
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP.
How do you trigger the do skill with a natural request?
You do not need a special command. The skill's description says it should be used when you ask the agent to execute, run, or carry out a plan, especially one created by make-plan. Ask in plain language and name the plan file or the plan you have in context.
Clear requests work best. Say which plan you mean, that you want it executed phase by phase, and whether you want commits and pushes handled as the skill describes.
- "Execute the plan in docs/plan.md using the do skill."
- "Carry out the phases of this plan one at a time, with verification before each commit."
- "Run the plan we made with make-plan."
What should you prepare before running it?
The skill needs a Git repository where the agent has push access to a working branch, and an agent runtime that can deploy subagents, such as Claude Code. A multi-phase plan works best, ideally one created by the companion make-plan skill, since the do skill is written to follow plan phases.
Use a working branch rather than a production or shared branch. The skill commits and pushes on its own once checks pass, so review what it will change before you start.
- A Git repository with push access to a working branch.
- A subagent-capable agent runtime, for example Claude Code.
- A phased implementation plan, ideally from make-plan.
- Network access, because the skill pushes to the remote.
What are the safety notes for scripts and network use?
According to the listing, this skill does not run scripts, but it does use the network. Its only file is SKILL.md. The network use comes from pushing to the working branch after each verified phase, which changes your repository history on the remote.
Before running it, confirm you are on the right branch, that your remote is the one you expect, and that you are comfortable with automatic commits and pushes. Review the commits after each phase if you want to catch problems early.
- No scripts are run by this skill (per the listing).
- Network is used for pushing to a working branch.
- Commits happen only after verification passes, but pushes still change shared history if the branch is shared.
What tips help you get reliable results?
Give the plan clear phases with checkable outcomes, because the verification subagent needs something specific to prove. Keep the plan's anti-patterns written down, since the anti-pattern subagent searches for the bad patterns the plan names.
For small jobs, skip this skill. The many subagents and commits add overhead that a simple one-off edit does not need.
- Write each phase with outcomes a checker can confirm.
- List known bad patterns in the plan so the anti-pattern check has targets.
- Use fresh subagents for phases with large or unclear context, as the source recommends.
What people use it for
- Executing a multi-phase feature plan one phase at a time with separate verification before each commit.
- Running a refactoring plan where anti-pattern checks search for patterns the plan names as bad.
- Carrying out a migration plan on a working branch with a progress report backed by command output and file changes.
- Handing a plan from make-plan to the agent so each phase starts with fresh subagents and plan context.
Frequently asked questions
How do I install the do agent skill?
Download the skill ZIP and unzip it so the skill folder with SKILL.md sits in ~/.claude/skills/ for personal use or in .claude/skills/ inside a project. In the Claude apps, you can upload it under Settings → Capabilities → Skills.
What does the do skill need to run?
It needs a Git repository with push access to a working branch and an agent runtime that can deploy subagents, such as Claude Code. The listing says it does not run scripts but does use the network.
Is the do skill free?
The source gives its licence as Apache-2.0. It does not state pricing, so check the source repository for any terms beyond the licence.
Will the do skill commit to my main branch?
The skill is designed to commit only after verification passes and to push to a working branch after each verified phase. Because it commits and pushes on its own, the listing advises against using it on production or shared branches without review, so use a working branch.
When should I not use the do skill?
Skip it for simple one-off edits, since many subagents and commits add overhead you do not need. Also avoid it on production or shared branches without review.
What is the difference between do and make-plan?
According to the source, make-plan creates a plan and do executes a phased implementation plan using subagents. The do skill is especially suited to plans made by make-plan, though it can run other plans that have clear phases.
New agent skills, repos and AI tools — every week
The best new skills and open-source projects, the fastest-growing repos, and one free prompt. Five minutes to read. Unsubscribe anytime.
Our verdict
The do skill suits developers with multi-phase plans who want disciplined, checked execution with evidence-backed reports, especially in Claude Code. It is a poor fit for small edits or for work on production or shared branches where automatic commits and pushes are risky without review.
Download the skill