Guide · Agent skill
How to use the Inline Plan Executor agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The Inline Plan Executor (the executing-plans skill from obra/superpowers) tells a coding agent to carry out a written implementation plan itself, task by task, in the current session, instead of handing each task to a separate helper agent. This guide explains what the skill makes the agent do, how to install it, how to trigger it, and what it needs before you rely on it. By the end you will be able to decide whether it fits your project and start a run safely.
Inline Plan Executorobra/superpowers · MIT Download the skill Quick start
- 1Download the ZIP. Download the executing-plans ZIP from this page.
- 2Unzip into a skills folder. Unzip into ~/.claude/skills/ for personal use, or into the project's .claude/skills/ folder.
- 3Check the requirements. Confirm Git, the superpowers skills and a shell that can run the sdd-workspace script are available.
- 4Prepare your plan. Make sure the plan is complete, detailed and mostly independent tasks, ideally written with superpowers:writing-plans.
- 5Ask for inline execution. Name the plan file and ask the agent to execute it in this session without helper agents per task.
- 6Review the ledger and final review. Read progress.md and the rulings, then check the final whole-branch review before any merge or push.
What the skill makes the agent do
The skill is for the case where you already have a detailed implementation plan for a coding project and want the agent to build it in one session. The agent works through the plan in order, writes tests first, checks each step's output against what the plan says to expect, and keeps a ledger file as a record of progress.
At the end, the agent runs one fresh-context review of the whole branch. The skill also sets rules for small problems: the agent records its own decisions in the ledger and keeps going, and it stops to ask you only for a short list of cases, such as irreversible or destructive operations, security-sensitive actions, and side effects outside the worktree like merges, pushes or publishing.
- Works through the plan task by task without a helper agent per task.
- Writes tests first and confirms each one fails, then passes.
- Logs every ruling in a progress ledger, so you can trace what was decided.
- Runs one final review of the whole branch, then uses the finishing-a-development-branch skill.
Install the skill
Download the ZIP from this page and unzip it. The skill is a folder containing a SKILL.md file. Put it in one of two places: your personal skills folder at ~/.claude/skills/, or a project's .claude/skills/ folder if you want it available only in that project.
If you use the Claude apps instead of the file system, you can upload the skill under Settings → Capabilities → Skills.
- Personal install: unzip into ~/.claude/skills/.
- Project install: unzip into the project's .claude/skills/ folder.
- Claude apps: upload under Settings → Capabilities → Skills.
Check the requirements first
The skill depends on Git and on several other skills from the superpowers set. It uses git worktrees and commits, so your project must be a Git repository. The skill also refers to a shared script, sdd-workspace, that is not included in the file you download, so you need a shell that can run that script from the superpowers setup.
Confirm these before starting a run, because a missing dependency will stop the agent partway through the plan.
- Git installed, with the project as a repository.
- superpowers:using-git-worktrees, superpowers:writing-plans, superpowers:subagent-driven-development, superpowers:finishing-a-development-branch and systematic-debugging available.
- A shell able to run the sdd-workspace script referenced in the Setup section.
- The skill runs scripts and does not use the network.
Trigger the skill with a natural request
You do not need a special command. The skill's description says to use it when executing an implementation plan in the current session as the implementer, either because you chose inline execution or because no subagent tool is available. Make that clear in your request and point the agent at your plan file.
The source does not list exact phrasings, so a plain request that names the plan file and says to execute it in this session is the safest approach.
- Example request: "Execute the plan in docs/plan.md inline in this session."
- Say explicitly that you do not want helper agents per task if your tool supports them.
- Ask the agent to resume from the ledger if a previous run was interrupted.
Use it on a plan that is ready
The skill assumes the plan is fully written and detailed. A vague plan makes the agent guess, and the skill is built around not guessing. Plans written with the superpowers:writing-plans skill fit this best. Tasks should be mostly independent.
Long plans need care. If the agent's context may be compacted, later tasks get less attention. The ledger lets a run resume, but the skill itself suggests a different approach (superpowers:subagent-driven-development) when the plan is long or when you want a review gate on every task.
- Good fit: a complete plan from a planning skill, one session, traceable decisions.
- Poor fit: no written plan, or a long plan where later tasks matter most.
- Keep merges, pushes and publishing under your own review; the agent is told to stop and ask, but the rule depends on the agent following it.
Read the ledger and the workspace
Each plan gets its own git-ignored workspace directory under .superpowers/sdd/. The ledger lives at progress.md inside it and starts with a line naming the plan file. Lines reading Task N: complete mark finished tasks, so the agent resumes at the first unfinished one after an interruption.
Rulings appear in the ledger in the form: Ruling: what was decided — why — what it costs if wrong. Read these after a run to check the agent's decisions.
- Workspace path: <repo-root>/.superpowers/sdd/<plan-basename>/.
- Ledger: progress.md, with the plan file named on the first line.
- Warning: git clean -fdx deletes the workspace; recover from git log if that happens.
# SDD ledger — plan: <plan file path>What people use it for
- Implementing a fully written feature plan from a planning skill in one session.
- Running a plan in a tool that has no subagent feature, so the plan runs directly.
- Running a plan unattended and reviewing a traceable ledger of decisions afterwards.
- Resuming an interrupted run by reading the ledger and git log.
Frequently asked questions
Is the executing-plans skill free to use?
The source states the skill is licensed under MIT. It does not list prices, so check the repository for anything beyond that licence.
What do I need to install the executing-plans skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ or a project's .claude/skills/ folder, or upload it under Settings → Capabilities → Skills in the Claude apps. You also need Git and the related superpowers skills.
Does the executing-plans skill need the internet?
The source says it does not use the network. It does run scripts, specifically the sdd-workspace script, which is not included in the file you download.
Can I use this skill without a written plan?
The source says no. The skill assumes a written, detailed plan exists, and a vague plan makes the agent guess. Write the plan first, ideally with superpowers:writing-plans.
How does this compare with subagent-driven development?
Inline execution runs every task in one session with a single final reviewer. The source says superpowers:subagent-driven-development is preferred when you want a review gate on every task or the plan is long enough that later tasks would run on compacted context.
What should I do if the agent stops and asks me a question?
The agent should stop only for irreversible or destructive operations, security-sensitive actions, side effects outside the worktree such as merges or pushes, or a plan so broken that every path is a guess. Answer the question, then let the run continue from the ledger.
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
Use this skill if you have a complete written plan, want it built in one session at lower cost, and want a clear ledger of decisions. Pick superpowers:subagent-driven-development instead if you want a review on every task, your plan is very long, or you plan to merge or publish without reviewing each step yourself.
Download the skill