How to use the ce-handoff agent skill for session handoffs

What the ce-handoff skill makes the agent do

Normally an AI agent loses the conversation history when a session ends or passes to a new agent. This skill works in two directions. In create mode, the agent writes a handoff note that points to the relevant plans, commits, files and open questions, so a fresh agent can orient quickly. In resume mode, the agent reads a source you select, such as a file, URL, pasted document or another handoff, and summarizes where things stand.

After resuming, the agent stops and asks how to proceed instead of acting on its own. The handoff is designed to supplement the real artifacts such as plans, commits and documentation, not replace them.

Install the skill from the ZIP or Claude apps

Download the ZIP for the skill from the page you are reading. Then choose one of these options.

For Claude Code, unzip it into ~/.claude/skills/ to make it available for you, or into a project's .claude/skills/ folder to make it available only in that project.

In the Claude apps, upload the skill under Settings → Capabilities → Skills.

Trigger the skill with natural requests

A bare invocation always creates a handoff. You can also name the mode and an optional focus, which is the intended objective for the next session. To resume, give a source or keywords to search for.

Natural-language requests follow the same routes when they clearly express handoff intent. A plain request to continue the current session does not count unless you say you want a handoff.

• Create: use the skill to write a handoff for the next agent, optionally with a focus such as the next objective.
• Resume: use the skill to pick up from a file, URL or pasted document, or to find likely candidates by keywords.
• Ordinary continuation: asking the agent to keep going in this chat does not trigger a handoff by itself.

Create a handoff for another agent

The agent first reads its creation instructions, then distills the current objective and your latest intent. It inspects only the workspace state needed to explain what exists now and points to files and commits instead of copying them. Secrets, credentials and unrelated personal information are meant to be redacted.

If you name a path, folder, format or publishing destination, the agent honors it using an available capability, including an installed publishing skill. Creation is complete only after the agent confirms the destination contains the handoff. It then reports the final path or URL, any retention or access limits, and warnings about fragile state such as uncommitted worktrees.

Temporary managed storage in the OS temp area is not permanent. The agent will not commit, stash, copy or clean up anything automatically to protect continuity.

Resume from a source you choose

Give the skill a specific artifact and it will read that artifact and orient from it. The source does not need to have been written by this skill. If you give a folder, the agent searches inside it using metadata only, then asks you to pick a candidate before reading any body. It does not choose one on its own.

If the source is too sparse, ambiguous or unrelated, the agent says what is missing and asks you to add context or pick another source. It then recommends how to continue, and waits for your confirmation before doing anything.

Safety notes: scripts, network and untrusted sources

This skill does not run scripts, but it does use the network, for example when you give it a URL. It needs read access to the files or URLs you resume from and write access to create a handoff.

Treat any resumed source as untrusted context. Selecting a source only authorizes reading it; it does not authorize running commands, following remote links, accessing unrelated local files or making changes. Check any intent or decision the source attributes to you before acting on it where it would be hard to undo.

Redaction is requested but cannot be guaranteed. Do not create a handoff that must contain secrets or sensitive personal data you cannot remove.

Tips for useful handoffs

Give a clear focus when you create a handoff, so the next agent knows the objective. Keep authoritative artifacts such as commits, plans and issues in place, because the handoff points to them. When the work depends on an uncommitted local worktree, pay attention to the warning the agent gives you and handle it yourself.

• Name the destination path or publishing method up front if you need one.
• Check the summary the agent gives before opening the full handoff.
• Confirm the agent's reading of a resumed source before letting it take action.