How to use the Compound Engineering Help agent skill (ce-help)

What the ce-help skill makes the agent do

CE's loop is ce-brainstorm, ce-plan, ce-work, ce-simplify-code, ce-code-review, and then ce-compound. The last step captures what you learned, and the next round's brainstorm and plan read it back. ce-help places your current work on this loop.

When you ask which skill fits, what comes next, how setup or configuration works, what things cost, how upgrades work, or why a skill misbehaved, the agent checks your project state. It looks at the conversation, the current branch, uncommitted changes, any open pull request, and recent plans. Then it recommends one next step and says why it fits.

• It recommends commands; it does not run them.
• It does not change any files.
• Git checks are read-only.
• It fetches public CE guide pages when an answer depends on details it cannot confirm from its own references.

Install the ce-help skill

You need the Compound Engineering plugin installed in a supported agent host. The skill itself installs in one of two ways. Download the ZIP for ce-help from this page and unzip it either into ~/.claude/skills/ to make it available across your projects, or into a project's .claude/skills/ folder to limit it to that project.

If you use the Claude apps, you can instead 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: Settings → Capabilities → Skills, then upload.
• Start a new session after installing, because hosts load skills at session start.

Trigger it with a natural request

The skill is set to run only when you invoke it, so start your message with the skill name. In Claude Code the default form is /ce-help followed by your question. The skill also accepts a question as its argument, described as your question about Compound Engineering.

You do not need special wording after that. Describe your situation in plain language and the agent will route it to the right reference.

Examples of questions it handles well

The skill is most useful when you are unsure which step comes next. It can also explain setup, cost, and upgrades, or help you work out why a skill did something unexpected.

If your message asks for the work itself, such as fixing a flaky test, the agent will invoke the skill that fits and do the work. Some skills are manual-only, so for those it gives you the exact command to type.

• Which skill fits a job, such as turning a rough idea into requirements.
• What comes next after a plan is written.
• Why a CE skill stopped or behaved differently than expected.
• What a setup or configuration option does, and whether it changes cost.
• How to update CE without running an old copy of a skill.

What it needs and what it does not do

ce-help needs three things to work fully: the Compound Engineering plugin installed in a supported host, a Git repository for project-state checks, and network access to fetch public documentation pages when needed. Without network access it answers from its own references and tells you when it could not check a page.

It does not run scripts. It does use the network, and it warns that its own recall of Compound Engineering may be out of date, so it checks public guide pages for details such as flags, config keys, and defaults.

• Requires: Compound Engineering plugin, a Git repository, network access for documentation pages.
• Runs scripts: no.
• Uses network: yes, to fetch public CE guide pages.
• Git commands are read-only.
• Does not change files or start CE workflows on your behalf unless you asked for the work.

Safety notes on scripts and network access

Because the skill runs no scripts, the main thing to know is its network use. When an answer depends on current behaviour, it fetches public pages from the Compound Engineering repository on GitHub. Those pages can be newer than your installed copy, so the skill tells you when an upgrade may be needed.

Since cost depends on which skills you run, the skill is careful not to start expensive runs for you. Many CE skills dispatch several agents or call other model providers, and the skill leaves that decision to you.

• Read-only Git checks only.
• No file changes.
• Public documentation pages are fetched when needed.
• Expensive multi-agent runs are recommended, not started.

Tips for getting useful answers

Say what you are trying to do right now: start something new, continue work in progress, find a skill for one job, fix a problem, or change CE's defaults. If the skill is unsure, it will ask this question before recommending anything.

If you are setting up a project, run ce-setup in it. It reports optional tools and can offer to create a repo config file. A missing optional tool is not a failure; only the workflows that use it need it.

• Name your goal and the state of the work, such as whether a pull request is open.
• Start a new session after any CE install or update.
• If a skill seems missing after an update, refresh the marketplace before updating the plugin, because updating alone can read a stale snapshot.
• Ask for the exact command when you want to run it yourself.