Guide · Agent skill
How to use the Execute Approved Tasks agent skill
Updated · written with AI from the project's own documentation and checked against our catalogue data
The execute skill is a workflow file for an AI coding agent. Once you have a task that is clearly defined, it tells the agent to build it in small verified steps and then report what changed, what was tested, and what is still open. This guide explains what the skill does, how to install it, how to ask for it, and where it fits and does not fit.
Execute Approved TasksYeachan-Heo/oh-my-claudecode · MIT Download the skill Quick start
- 1Download the ZIP. Download the skills/execute ZIP from this page and unzip it to get SKILL.md.
- 2Choose an install location. Place the folder in ~/.claude/skills/ for all projects, or in .claude/skills/ inside one project.
- 3Or upload in the Claude apps. In the Claude apps, go to Settings → Capabilities → Skills and upload the skill.
- 4Check prerequisites. Confirm your agent runtime supports SKILL.md skills and sub-agents, and that your project's build and test tools work.
- 5Describe a clear task. Ask for a specific build with its goal and constraints, such as which tests must pass.
- 6Review the completion report. Check the files changed, the evidence it works, and the open items before accepting the work.
What the execute skill makes the agent do
The skill covers the build phase of a task. When the work is understood, the agent confirms the task is clear enough to build, splits it into independent units, and runs genuinely independent units in parallel where that makes sense. Each unit gets the smallest correct change, reusing existing utilities and patterns, and the agent verifies as it goes rather than only at the end.
If the task is not clear, the skill tells the agent to plan first instead of building.
- Confirms scope before building
- Splits work into independent units and runs them in parallel where sensible
- Keeps changes small and reversible
- Prefers deleting code over adding it when behaviour is preserved
- Does not add dependencies unless you explicitly ask
- Treats placeholder TODOs, skipped tests and stub tests as blockers
Install the execute skill
Download the ZIP for skills/execute from this page and unzip it. The skill is a single file, SKILL.md. You can install it in one of two places:
For Claude Code, put the unzipped folder in ~/.claude/skills/ to make it available across your projects, or in a project's .claude/skills/ folder to limit it to that project.
In the Claude apps, upload the skill under Settings → Capabilities → Skills.
- Personal install: ~/.claude/skills/
- Project install: .claude/skills/ inside your project
- Claude apps: Settings → Capabilities → Skills
Check what the skill needs
The skill needs an agent runtime that supports SKILL.md skills and sub-agents. The source names the oh-my-claudecode plugin as one example, because it provides the executor, review and verify agents that this skill refers to. The project's own build and test tooling is also required, since the skill verifies against whatever the project already uses.
The skill does not run scripts and does not use the network. The only file in the package is SKILL.md.
- Agent runtime with SKILL.md and sub-agent support
- Example: the oh-my-claudecode plugin (provides executor, review and verify agents)
- Your project's existing build and test tools
- Runs scripts: no
- Uses network: no
How to trigger the skill with a natural request
Ask for the build directly and say what you want built. The skill fits best when the task is already defined. Include the goal and any constraints, such as which tests must pass or which files should not change.
The source also says that several other workflows route to this skill: autopilot, ralph, ultragoal, ultrapilot, pipeline and swarm. If you use one of those, execution is handled by this skill's workflow.
- Good: "Add rate limiting to the /login endpoint using the existing middleware pattern, and make sure the auth tests pass."
- Good: "Implement the three approved changes in the ticket, in parallel where they do not touch the same files, and report what you verified."
- Weak: "Look into how we could improve performance." Use a planning request first.
Choose the right amount of machinery
The skill scales its approach to the task. A single unit is implemented directly, then verified. Several independent units are delegated to executor agents in parallel. Long or unbounded work is tracked in a durable task list until the list is empty. Work that needs coordinated parallel workers uses team.
The skill also says not to set up coordination for work one focused pass would finish.
- Single unit: implement, verify, done
- Several independent units: executor agents in parallel
- Long-running or unbounded: durable task list until empty
- Coordinated parallel workers: team
Read the completion report
Before claiming done, the agent should have no pending tasks, tests passing or failures reported plainly, and verification evidence collected. The final output lists the files changed, what was implemented, the evidence it works, and what is still open.
The skill separates authoring from approval. It does not approve its own work; it hands off to review or verify.
- Files changed
- What was implemented
- Evidence it works
- What is still open
Safety notes
The skill itself runs no scripts and makes no network calls. It does, however, modify files and run tests, because the workflow depends on both. The skill also tells the agent to read the source of any skill or protocol before asserting what it does or skipping a step.
Review the diff before accepting it. Use the review or verify step the skill hands off to, rather than relying on the agent's own sign-off.
- No scripts, no network access in the skill itself
- Modifies project files and runs tests as part of the workflow
- Does not self-approve; hand off to review or verify
What people use it for
- Implementing a set of approved changes that touch separate files, run in parallel
- Finishing a single well-defined feature with tests run as it goes
- Carrying a long task list through to empty with a durable checklist
- Producing consistent completion reports that list changes, evidence and open items
Frequently asked questions
Is the execute skill free?
The source does not state a price or licence terms for the skill beyond the repository licence, which is MIT. Check the repository for any further terms before you rely on it.
What do I need to install the execute skill?
You need an agent runtime that supports SKILL.md skills and sub-agents. The source gives the oh-my-claudecode plugin as an example, since it provides the executor, review and verify agents the skill refers to.
Where do I put the execute skill folder in Claude Code?
Put it in ~/.claude/skills/ to use it across projects, or in .claude/skills/ inside a single project. In the Claude apps, upload it under Settings → Capabilities → Skills.
Does the execute skill run scripts or use the internet?
No. The source says it runs no scripts and uses no network. It does modify files and run tests as part of its workflow.
When should I use a planning skill instead of execute?
Use planning first when the request is vague or exploratory. The execute skill tells the agent to plan if the task is not clear enough to build, so it adds little value for those requests.
Can the execute skill approve its own work?
No. The source says authoring and approval are separate passes, and the agent should hand off to review or verify rather than self-approve.
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 the execute skill if you already know what you want built and want the agent to finish it with tests and a clear report of what changed and what is still open. Pick something else if your request is vague, you need an independent reviewer from the same tool, or your project does not allow the agent to run tests or edit files.
Download the skill