Guide · Agent skill
How to use the Polish Feature Live agent skill (ce-polish)
Updated · written with AI from the project's own documentation and checked against our catalogue data
The ce-polish skill is an agent skill from the EveryInc compound-engineering-plugin that helps you refine the look and interaction of a feature that already works, before you ship it. It starts your app's dev server, opens the page, and lets you say what could be better while the agent makes focused fixes that hot reload shows immediately. This guide explains what the skill does, how to install it, how to trigger it, and what it needs and risks.
Polish Feature LiveEveryInc/compound-engineering-plugin · MIT Download the skill Quick start
- 1Download the ZIP. Download the ce-polish skill ZIP from this page.
- 2Install the skill. Unzip it into ~/.claude/skills/ or a project's .claude/skills/, or upload it under Settings → Capabilities → Skills in the Claude apps.
- 3Check out a feature branch. Switch to the feature branch in a git repository, not the default branch.
- 4Ask for a polish pass. Request polish in plain language, optionally naming a PR number or branch.
- 5Choose live or traditional. Pick live mode, which needs an OpenAI API key, or traditional mode where you type your observations.
- 6Give feedback on the running page. Describe what could be better and let hot reload show each fix.
- 7Close the loop. Say you are done so the agent saves the changes with ce-commit and reports the commits and server URL.
What is an agent skill and what does ce-polish do?
An agent skill is a folder with a SKILL.md file and optional scripts and reference notes that gives an AI coding agent a specific, repeatable workflow. The ce-polish skill covers one job: polishing a working feature through user-directed live browser feedback.
You drive the review. The skill follows your observations rather than a generated checklist, so it is for UX refinement, not automated QA. The agent edits the in-scope code, hot reload updates the running page, and when you finish, the changes are saved as local git commits.
Install the ce-polish skill
Download the ZIP of the ce-polish skill from this page. You have two options.
For Claude Code, unzip it into your user skills folder ~/.claude/skills/ for all projects, or into a project's .claude/skills/ folder to share it with that repository.
In the Claude apps, open Settings, go to Capabilities, and upload the skill under Skills.
- Personal install: unzip into ~/.claude/skills/
- Project install: unzip into the project's .claude/skills/
- Claude apps: Settings → Capabilities → Skills, then upload the ZIP
Check the requirements before you start
The skill works on a git repository with a feature branch checked out. It cannot work on the default branch. It also needs a local dev server for your app; the skill detects common setups such as Next, Nuxt, Vite, Rails, SvelteKit, Astro, Remix, or a Procfile.
A browser capability in your agent harness is optional. Without one, you describe what you see and the agent works from your description.
Live mode has extra requirements: an OpenAI API key in the environment and the riffrec package, which the skill installs as a setup commit. The skill also depends on the ce-commit skill to save commits.
- Git repository with a working feature branch
- Local dev server for the app
- Optional: browser capability in the agent harness
- Live mode only: OpenAI API key in the environment and the riffrec package
- The ce-commit skill, used to save commits
Trigger the skill with a natural request
The skill is invoked by asking for it in plain language once it is installed, for example when a feature works but needs visual or interaction polish. It accepts an optional argument: a PR number, a branch name, or nothing to use the current branch.
When it starts, it asks whether you want live or traditional mode before any server starts. In traditional mode you type what could be better. In live mode you talk and draw on the page while a voice interviewer turns your speech into fix requests.
- Polish the feature on this branch before I merge.
- Polish PR 123 and start the dev server so I can look at the UI.
- Use the polish skill on the checkout-form branch; I'll describe what feels off.
Work through the polish loop
After the server is running, the skill tells you where it is and asks what could be better. It does not start a review pass while you are browsing. For each change you request, the agent inspects only as needed, edits the in-scope code, and lets hot reload update the page.
When you say you are done, the agent saves the polish changes with ce-commit and reports the commit or commits, the still-running server URL, and any residual blocker. In live mode it also lists any units that were not applied and the path of the session log.
- Tell the agent what to change, one observation at a time.
- Ask it to inspect the result if a browser capability is available.
- Say you are done to close the loop and receive the commit summary.
Safety notes: scripts and network use
The skill runs scripts and uses the network. The scripts include port detection and dev-server helpers. In live mode, the page sends screen content and your voice to an OpenAI-backed interviewer and to a local endpoint the skill runs, and it adds the riffrec dependency and a provider mount as a setup commit that stays after the session.
The skill never pushes and never opens a pull request, so you must push yourself. Avoid live mode on screens that show sensitive information.
- Live mode sends screen content and voice to an OpenAI-backed interviewer.
- The riffrec setup commit remains on the branch after the session.
- Traditional mode does not use the voice interviewer.
What people use it for
- Refining spacing, alignment and wording on a feature before merging its pull request.
- Letting a designer or product person describe UX problems on a running page and have them fixed in code.
- Reviewing interaction details such as button states or form behaviour on a feature branch.
- Using live mode to talk through and draw on screen issues while the interviewer converts them into fix requests.
Frequently asked questions
How do I install the ce-polish agent skill?
Download the ZIP from this page and unzip it into ~/.claude/skills/ for personal use or into a project's .claude/skills/ folder. In the Claude apps, upload the ZIP under Settings → Capabilities → Skills.
What do I need to use ce-polish?
You need a git repository with a feature branch, a local dev server for your app, and the ce-commit skill. A browser capability is optional, and live mode additionally needs an OpenAI API key and the riffrec package.
Is ce-polish free to use?
The source is published under the MIT licence. This page does not state pricing for any service, and live mode uses OpenAI, so check that provider's terms for any costs you incur.
Can ce-polish find bugs automatically?
No. The skill follows only your observations and does not generate a checklist, so it is not a substitute for autonomous QA or full testing.
Does ce-polish push my code or open a pull request?
No. It edits and saves changes as local commits and never pushes or opens a PR, so you need to push the branch yourself.
Why won't ce-polish work on my main branch?
The skill is explicitly forbidden from working on the default branch. Create or check out a feature branch first, then run the skill.
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 ce-polish if you have a working feature on a feature branch and want focused, hands-on UI refinement with a second pair of eyes before merge, especially if you can describe problems but not fix them in code. Pick something else if you need automated bug finding or full testing, or if you cannot send screen content and voice to OpenAI for live mode.
Download the skill