Guide · Agent skill

How to use the Feedback Sweep agent skill (ce-sweep)

Updated 2026-10-29 · written with AI from the project's own documentation and checked against our catalogue data

Feedback Sweep (ce-sweep) is an agent skill that checks the feedback sources you configure, such as Slack channels and GitHub Issues, for items that are new since its last run. It acknowledges those items at their source, analyzes attached recordings, checks whether claimed fixes have reached the default branch, and writes a plan file you can hand to the lfg workflow. This guide is for readers who have not used agent skills before, and it explains how to install the skill, start a run, and what to check before you let it post anything.

Feedback SweepEveryInc/compound-engineering-plugin · MIT Download the skill

Quick start

1Download the ZIP. Download the ce-sweep skill ZIP from this page.
2Unzip into a skills folder. Unzip it into ~/.claude/skills/ for personal use, or into the project's .claude/skills/ folder. Alternatively, upload the ZIP in the Claude apps under Settings, then Capabilities, then Skills.
3Confirm your requirements. Make sure git, the gh CLI and Python 3 are available, and that you have Slack and GitHub access for the sources you want to sweep.
4Run first-time setup. Type /ce-sweep setup in an interactive session and answer the questions about your feedback sources and which sources are approved for source-side writes.
5Run a sweep. Type /ce-sweep to fetch new items, acknowledge them at approved sources and update the plan.
6Review the plan. Open the plan file and check item content and any sensitive flags before sharing it.
7Run lfg on the plan. Use the lfg command shown in the run summary on the plan file to turn it into work.

What the skill makes the agent do

When you run ce-sweep, the agent works through a fixed sequence. It fetches new items from each configured source, acknowledges them at the source, analyzes any attached screen recordings or audio into transcripts and notes, and checks whether any fix an item claims has been merged to the default branch. Open items are then folded into a rolling plan file.

The skill keeps its own state file so that later runs do not repeat work already done. Item content and media from feedback are stored in the repo-internal state and plan, so review them before sharing the repository.

Sources named in the skill description: Slack, GitHub Issues, and email (marked experimental).
Output: a rolling plan at docs/feedback-sweep/plans/feedback-sweep-plan.md by default, or under your configured docs_root.
Raw media is never committed. Only the plan and the repo-internal state are.

Install the ce-sweep skill

Download the skill ZIP from the page you are reading. Then choose one of two install locations. For a personal install available in all projects, unzip it into ~/.claude/skills/. For a single project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the ZIP under Settings, then Capabilities, then Skills.

Personal: ~/.claude/skills/
Project only: .claude/skills/ inside the project folder
Claude apps: Settings → Capabilities → Skills, then upload

Set up feedback sources on the first run

The first run asks you to configure sources. The setup interview is interactive only, so run it yourself rather than from a scheduled job. The answers are written to .compound-engineering/config.local.yaml in your repository.

Each source has its own entry. A source only receives writes to its own service (acknowledgements or close-outs) when its entry is approved. Sources that are not approved still have their items fetched and recorded, but they receive no source-side writes.

Config keys include feedback_sources and sweep_* settings, each with a default.
Unapproved sources: items are recorded as ack_deferred and nothing is posted.
Email is experimental and the skill description does not say it works reliably.

Trigger a sweep

The skill is set to manual invocation, so the reliable way to start it is the slash command. Type /ce-sweep to run a sweep, /ce-sweep setup for first-time setup, or /ce-sweep reconfigure to change sources. On Codex, the dollar-prefixed form $ce-sweep is used instead. Plain-language requests are not documented as a trigger in the source.

For unattended runs, add the non-interactive token. The skill accepts mode:non-interactive, and the older alias mode:headless is also recognized.

/ce-sweep: run a normal sweep
/ce-sweep setup or /ce-sweep reconfigure: change configuration
/ce-sweep mode:non-interactive: run without prompts, for scheduled use

/ce-sweep mode:non-interactive

Read the result and hand the plan to lfg

At the end of a run the skill prints a summary that ends with the plan path and a line to run the lfg workflow on that plan. Review the plan before running lfg, because it is built from item content that people wrote.

If a run stops early, the summary records why. A run that finds the lease held by another run is recorded as aborted-locked and exits without writing.

Check the summary for the plan path.
Review the plan and any sensitive flags before sharing it.
Run the lfg command the summary gives you on the plan file.

/lfg docs/feedback-sweep/plans/feedback-sweep-plan.md

What people use it for

A product team triages user reports posted in a Slack channel and gets an open-items plan after each sweep.
A maintainer runs recurring sweeps over GitHub Issues and checks whether reported fixes have been merged to the default branch.
A team that already uses the compound-engineering lfg workflow turns new feedback into a plan file for that workflow.
A scheduled non-interactive run keeps the plan current between manual sessions, for sources that are approved.

Frequently asked questions

How do I install the ce-sweep agent skill?

Download the skill 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, you can upload the ZIP under Settings, then Capabilities, then Skills.

What do I need to run ce-sweep?

The skill uses git and the gh CLI for GitHub Issues and fix verification, and Python 3 for its state script. You also need Slack and GitHub access for the sources you configure. The source does not confirm the exact auth method or which extra tools media analysis needs, such as ffmpeg or a transcription tool.

Is ce-sweep free and what is its licence?

The skill is listed under the MIT licence. The source does not state any price, so check the repository for current terms.

How do I start a sweep with a natural request?

The skill is set to manual invocation, so use the slash command /ce-sweep. The source does not document plain-language requests as a trigger.

Will the agent post replies to my Slack or GitHub without asking?

Only for sources whose config entry is approved. Unapproved sources get no acknowledgements or close-outs, but their items are still fetched and recorded. Review the approval settings during setup before your first sweep.

Why does my run say first run requires interactive setup?

No feedback sources are configured yet, and first-run setup needs an interactive session. Run /ce-sweep setup yourself once, then scheduled non-interactive runs can use mode:non-interactive.

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-sweep if your team triages feedback from Slack or GitHub Issues, you use the lfg workflow, and you are comfortable reviewing the plan and approving which sources the agent may write to. Pick something else if you need reliable email handling, want no agent writes to public channels, or handle sensitive feedback that you cannot store in the repository.

Download the skill