Source: mattpocock/skills · Maintainer: community: Matt Pocock · License: MIT · Checked: 24 October 2026
What is the handoff skill?
The handoff skill writes the conversation you are in to a single markdown file, saved in your OS temp directory, so a fresh agent can read it and continue the work somewhere else. You run it by typing /handoff, optionally followed by what the next session is for.
It comes from mattpocock/skills, which Matt Pocock describes in the README as "My agent skills that I use every day to do real engineering - not vibe coding." The skill's description reads: "Compact the current conversation into a handoff document for another agent to pick up."
As of 24 October 2026, its skills.sh page listed 1.0M installs (snapshot, 24 October 2026).
What does it do?
The SKILL.md gives the agent five instructions:
- Write a handoff document summarising the conversation so a fresh agent can continue.
- Save it to the OS temp directory:
$TMPDIR, else/tmp, or%TEMP%on Windows. Not the current workspace. - Add a "suggested skills" section naming the skills the next agent should load.
- Don't duplicate existing artifacts. Specs, plans, ADRs, issues, commits and diffs are referenced by path or URL, not copied.
- Redact sensitive information, such as API keys, passwords and personal data.
If you pass text after the command, the agent treats it as the focus of the next session and tailors the document to it. The frontmatter's argument-hint ("What will the next session be used for?") shows that prompt in autocomplete; argument-hint is a documented Claude Code skill frontmatter field.
Triggers. None automatic. The frontmatter sets disable-model-invocation: true, and agents/openai.yaml sets allow_implicit_invocation: false for Codex.
Folder contents: SKILL.md and agents/openai.yaml (display name "Handoff"). No scripts or templates.
When should you use handoff instead of /compact?
The handoff docs are direct about this: what the skill gives you is portability, not better compression. If nothing is moving, use Claude Code's built-ins. The docs name four situations where a file is the right tool:
| Situation | Why a file helps |
|---|---|
| Switching agents, such as Claude Code to Codex | The new agent cannot see the old context |
| Moving to another directory or repo | A prototype directory is the common case |
| Sending the work to a colleague | They need something they can read |
| Forking a side task found mid-task | You keep working while a second agent takes the fork |
Here is how it compares with the Claude Code commands, as the Claude Code commands reference describes them:
| Option | What it does | Output |
|---|---|---|
/compact |
Summarizes the conversation to free up context, optionally with focus instructions | Same session continues |
/clear |
Starts a new conversation with empty context | Nothing carried over |
/fork |
Copies the current conversation into a new background session | A copy of the conversation in a new background session |
/handoff (this skill) |
Writes a targeted summary for a stated next task | A markdown file you can move anywhere |
The docs say the fork case is the one people skip. You stay in your design conversation, hand a copy of the context to a second agent (often running the prototype skill in its own directory), get the answer back, and never re-explain anything.
How do you install the handoff skill?
The README offers a plugin or editable files, one per agent, since both together give you every skill twice.
Claude Code plugin (whole collection, updates itself). From Anthropic's official plugin marketplace:
claude plugin install mattpocock-skills@claude-plugins-official
Plugin skills are namespaced, so it appears as /mattpocock-skills:handoff.
Codex plugin. From the README:
codex plugin marketplace add mattpocock/skills
codex plugin add mattpocock-skills@mattpocock
Just this skill, as editable files. The command on skills.sh is:
npx skills add https://github.com/mattpocock/skills --skill handoff
Add -g to install it for all projects. For Claude Code, the folder lands in .claude/skills/handoff/ in your project, or ~/.claude/skills/handoff/ with -g, the standard Claude Code skill locations. Codex and Cursor read the same SKILL.md format; see Codex skills and Cursor skills. Since the main use is moving work between agents, installing it in both is reasonable.
claude.ai. Custom skills upload as a ZIP of the folder in Customize, then Skills (Claude Help Center). As shipped, this SKILL.md will not upload: the Claude Code skills docs say claude.ai uploads accept only name, description, license, compatibility, metadata and allowed-tools, and fail on other keys such as argument-hint and disable-model-invocation. The temp directory step also assumes a local file system, so it fits a coding agent better.
How do you use it?
Type /handoff and say what the next session will do. Then ask for the file path, open the new session, and point it at the file.
Example prompts (suggestions, not quotes from the repo):
/handoff Next session continues this refactor in Codex./handoff A second agent will build a throwaway prototype of the settings page in ./proto./handoff My colleague picks up the payment webhook bug tomorrow. Keep the reasoning for why we ruled out retries./handoff Fork: investigate the flaky upload test while I keep working on the API.
In the new session, something like Read /path/to/handoff.md, then continue is enough.
Requirements. None beyond a coding agent that can write files. No packages or keys.
Tips and limits
From the repo's docs, not from testing:
- Keep the path. The most reported problem is losing the file. Temp paths are long, so ask for the path before you move on.
- Temp is not durable. Some environments clear temp between sessions (the docs name Codex), and a reboot empties
/private/tmp. Copy the file somewhere durable if the next session isn't starting soon. - Point at the file, don't paste it. Interpolating the summary into a shell command such as
claude "<summary>"can silently truncate text that contains backticks or$(...). - Pass the argument. Without a stated purpose, the summary can capture the what but not the why.
- Read it before you hand it over. The next agent trusts the document. A guess written as a fact ("X is done") becomes a false premise, so downgrade anything that was only assumed.
- Long-lived facts go elsewhere. If something will still be true next month, it belongs in CLAUDE.md, not a handoff.
The skill runs no scripts and writes a single file outside your project. It is MIT licensed. The skills.sh audits showed Pass from Gen Agent Trust Hub, Socket and Snyk as of 24 October 2026.
Save it in your Kipt library
In Kipt, open Skills and paste the skill's GitHub folder link into the import box. Kipt reads the SKILL.md and its files; because this repo has many skills, it shows a list to pick from. You can also drop the folder or a .zip. Your library lives in your own private GitHub repo and installs into Claude Code or Codex with one command, which suits a skill whose job is moving between the two:
npx skills add https://github.com/<you>/<repo>.git --all -g
Kipt is not affiliated with Matt Pocock or mattpocock/skills.
Similar skills
- Prototype skill: the most common destination for a forked handoff.
- Grilling skill: settle the plan first, then hand off the result.
- Implement spec skill: when the work is already captured in a spec, reference it instead of summarising.
- Writing plans skill: Superpowers' skill for turning a spec into a task-by-task plan file.