Guide · Agent skill
How to use the ce-ideate agent skill: generate improvement ideas
Updated · written with AI from the project's own documentation and checked against our catalogue data
ce-ideate is an agent skill, a set of instructions and reference files that a coding agent loads when a request matches its purpose. This one is for generating improvement ideas for a project: it has the agent study the repository, gather context from several sources, critique every candidate idea, and keep only the survivors with reasons for each rejection. After this guide you will know what the skill does, how to install it, how to phrase a request that triggers it, and what it needs and touches while it runs.
Generate Improvement IdeasEveryInc/compound-engineering-plugin · MIT Download the skill Quick start
- 1Download and unzip the skill. Download the skill ZIP from this page and unzip it into ~/.claude/skills/ for personal use or the project's .claude/skills/ folder for one project.
- 2Or upload it in the Claude apps. In the Claude apps, open Settings, go to Capabilities, then Skills, and upload the skill.
- 3Confirm you are in a Git repository. Open your agent in the project. The skill needs a Git repository and runs git rev-parse --show-toplevel to find the root.
- 4Ask for ideas in plain language. Request ideas, improvements, or surprising directions for a named subject, optionally with a focus area or constraint.
- 5Answer any clarifying questions. If the subject is unclear, the skill asks up to a few questions. Choose a numbered option, or pick Surprise me, or Cancel to exit.
- 6Review the cost line and let it run. The skill shows a cost line before dispatching sub-agents. Confirm it, then wait while grounding and ideation run.
- 7Read the ranked ideation document. Open the saved document in your docs ideation folder, review the survivors and the rejection reasons, then choose one to take to ce-brainstorm.
What the ce-ideate skill makes the agent do
When you ask for ideas about a project, the skill makes the agent first study the repository. It then dispatches several helper sub-agents to gather grounding from the code, past learnings, issue trackers, Slack, and the web. The agent generates a full candidate list before judging any of it, critiques every idea, and keeps only survivors, explaining each rejection.
The result is a ranked ideation document saved to a docs folder. The skill then hands off to a separate brainstorming skill (ce-brainstorm) for choosing and shaping one idea. It does not write requirements, plans, or code.
- Grounds ideas in the repository before generating any
- Generates many candidates, then critiques all of them
- Keeps survivors only, with a stated reason for each rejection
- Saves a ranked ideation artifact under the docs root
- Points you to ce-brainstorm for action instead of planning on its own
Install the ce-ideate skill
Download the ZIP for the skill from the page you are reading. Unzip it into one of two locations, depending on how you want it available. A personal install goes in ~/.claude/skills/. A project install goes in the .claude/skills/ folder of a specific project, so it applies only there.
If you use the Claude apps instead, you can upload the skill under Settings, then Capabilities, then Skills, rather than placing files on disk.
- Personal: unzip into ~/.claude/skills/
- Single project: unzip into the project's .claude/skills/ folder
- Claude apps: Settings → Capabilities → Skills, then upload the skill
How to trigger the skill with a natural request
You do not need a special command. The skill's description says to use it when you want ideas, improvements, or surprising directions before choosing one to develop. Ask in plain language that names the subject, and optionally a focus area or constraint.
The skill's argument hint is a focus area or constraint in brackets, plus an optional output switch for markdown. Output is either HTML or markdown, never both in one run. Use a request that asks for ideas, not for a refinement of an idea you already have.
- 'What should we improve next in this repo?'
- 'Give me surprising directions for the onboarding flow, keep it under our current auth setup.'
- 'Generate improvement ideas for the CLI error handling, output:md'
- 'Surprise me with ideas for this codebase.'
What the skill needs and what it touches
The skill needs a Git repository. It runs git rev-parse to locate the repository root, which you can see yourself with the command below. It also needs the host agent to support sub-agents. The source does not list any API keys.
Slack and web research depend on which tools the host agent has available. The skill runs no scripts, but it does use the network, through web research and any other sources the host provides.
- Git repository required
- Sub-agent capability required in the host agent
- Slack and web research only when the host provides those tools
- No scripts are run
- Network access is used
git rev-parse --show-toplevelSafety notes before you run it
Because the skill uses the network, expect outside requests when web research is enabled in your host. Check what your host allows before running it on sensitive projects.
The skill reads learnings from a solutions folder under the docs root and writes its output under an ideation folder. A docs root can be set in .compound-engineering/config.yaml. If it is unset, the default is docs. The skill also shows you a cost line before it dispatches agents, so you can stop it there.
- Network use is expected; review host tool permissions first
- Output goes to the ideation folder under the docs root
- Read docs_root from .compound-engineering/config.yaml only
- If grounding fails, the skill warns you and continues
Tips for better ideation results
Name the subject clearly. The skill needs an identifiable subject. If it cannot find one, it asks a question, and it may stop if the topic is not about software and has no clear subject. Keep focus hints short. The skill asks no more than three clarifying questions, and it avoids asking about solution direction, audience, or success criteria, because ce-brainstorm covers those.
Once you have the ranked document, pick one idea and open ce-brainstorm to shape it. Do not jump straight to planning from the ideation output.
- Name the repository or feature area explicitly
- Use a focus hint to narrow the scope
- Pick one surviving idea and move it to ce-brainstorm
- Re-run with a different focus if the list feels generic
What people use it for
- Finding the next feature or fix to explore in an existing codebase, with the reasons behind each choice written down
- Triaging a backlog by asking which recurring issue themes are worth investment, drawing on the issue tracker and past learnings
- Generating surprising directions for a module, then keeping a written record of why the rejected ideas were dropped
- Preparing a shortlist of ideas to hand to ce-brainstorm for a team discussion
Frequently asked questions
How do I install the ce-ideate 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 instead upload it under Settings, then Capabilities, then Skills.
What does ce-ideate require to run?
It requires a Git repository, because it runs git rev-parse --show-toplevel to find the repository root, and a host agent with sub-agent capability. Slack and web research depend on the tools your host provides. The source lists no API keys.
Is the ce-ideate skill free?
The source does not state a price or licence terms beyond identifying the item as MIT licensed. Check the repository's licence file for the full terms before you rely on it.
Does ce-ideate run scripts or access the network?
It runs no scripts, according to the source. It does use the network, for web research and any other sources the host makes available.
How is ce-ideate different from ce-brainstorm?
ce-ideate generates and ranks ideas before you choose one. ce-brainstorm shapes one chosen idea. The skill's own description says to use ce-brainstorm when you already have an idea to refine, and ce-pov when judging one idea already on the table.
Why did the skill ask me a question or stop?
It asks when the subject is not identifiable, and it may stop when a non-technical topic has no clear subject, because it needs something concrete to ground ideas in. Name the project or feature area and try again.
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-ideate if you have an existing codebase and want grounded, reasoned improvement ideas with a written record of why each was kept or rejected. Pick something else if you already have an idea to refine, want to judge one idea on the table, or need help with a topic that has no clear subject, since those belong to ce-brainstorm, ce-pov, or a direct conversation.
Download the skill