claude-handoff

An agent skill by mattpocock, from mattpocock/skills. Tags: claude-code, handoff, multi-agent, workflow.

What it does

Hand the current conversation off to a fresh background agent that picks up the work immediately.

Install

With the skills CLI, which installs into Claude Code, Codex, Cursor and other agents:

npx skills add mattpocock/skills --skill claude-handoff

Or copy the skill folder into Claude Code's skills directory by hand (~/.claude/skills for every project, or .claude/skills inside one):

git clone --depth 1 https://github.com/mattpocock/skills
cp -r skills/skills/in-progress/claude-handoff ~/.claude/skills/claude-handoff

Safety box score

Grade C, moderate risk: 30 of 100 risk points, rated 2026-10-16. The grade rates what the skill and its scripts can reach on the machine of whoever installs it, across eight categories from shell execution to secrets access. How the score works.

This skill hands the current conversation to a new background Claude agent by writing a summary to temp storage and launching `claude --bg` with that summary as the prompt. The main risks are that it automatically sends the handoff content off-machine to the Claude service, and that the summary can carry any mistakes in redaction or scoping into the next agent.

Category levels, from 0 (none) to 3 (high)
Category	Level	Why
Code & shell execution	2 (moderate)	It explicitly tells the agent to run a shell command, `claude --bg --name ... -- "$(cat <summary file>)"`, which starts a new process from the command line.
Network access	2 (moderate)	Launching `claude --bg` sends the handoff summary to the Claude service, so conversation content leaves the machine as part of the prompt.
Filesystem reach	1 (low)	It writes a handoff summary to the OS temporary directory and uses the current working directory for the background agent, but does not describe broader file access.
Secrets & credentials	1 (low)	The skill warns to `Redact any sensitive information` because the summary becomes the agent's prompt, but it does not directly read credential stores or key files.
Privilege & persistence	0 (none)	No sudo, service registration, global installs, or OS/shell setting changes are described.
Instruction hijack surface	0 (none)	The quoting note is defensive (`Passing the file keeps the shell from running backticks or expanding claude-handoff: Claude Code skill by mattpocock · Skills Explorer Skills Explorer Search skills Add Skills ); it does not instruct the agent to ignore rules or fetch instructions elsewhere.
Irreversible actions	2 (moderate)	It automatically launches a background agent that continues work immediately and returns without a human review step.
Transparency	0 (none)	The skill files are plain text YAML/Markdown with no binaries, hidden code, or unpinned downloads.

Before installing

Read `SKILL.md` and confirm you are comfortable with the summary being sent to a background Claude session.
Test once in a non-sensitive conversation to verify the temp-file workflow and quoting behave as expected.
Check that your usual handoff summaries do not include secrets, personal data, or unreleased material.
If your environment has strict data controls, confirm that `claude --bg` is permitted to send prompts off-machine.

Rated 2026-10-16 by an automated review of 2 files from the skill's repository.

Source

Repository: mattpocock/skills (all skills from this repository)
Path: skills/in-progress/claude-handoff/SKILL.md
Branch: main
Updated: 2026-10-05

handoff — Compact the current conversation into a handoff document for another agent to pick up.
continual-learning — Orchestrate continual learning by delegating transcript mining and AGENTS.md updates to `agents-memory-updater`.
orchestrate — Use only when the user explicitly types `/orchestrate <goal>` to decompose a large task.
principle-guard-the-context-window — Apply when context is filling up: large outputs, long files, repeated reads, fan-out planning.
ralph-loop — Start a Ralph Loop for iterative self-referential development.
swarm — Fan out N parallel workers, drain them, and return one report. Use for /swarm, 'swarm this', or parallel coverage, races, gauntlets, and exploration.