Claude Handoff Control — Development skill for Claude Code
Stop hook for Claude Code: returns the turn when the agent hands it back with an excuse instead of doing the work.
How to install Claude Handoff Control
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open mkdir-username/claude-handoff-control and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Claude Handoff Control does
Stop hook for Claude Code: returns the turn when the agent hands it back with an excuse instead of doing the work.
Alternatives in Development
- Taskmaster — Stop hook for Claude Code that keeps the agent working until all plans and user requests are 100% complete 494 ★
- Nonstop — Autonomous work mode for Claude Code 263 ★
- CoderBar — Native macOS agent control for Codex and Claude — monitor, approve, ask, and jump back from the menu bar 100 ★
README
claude-handoff-control
Made by and for developers tired of answering the obvious.
A coding agent keeps handing the turn back with questions it could settle itself. Real last lines from my judge log, translated from Russian:
- "Fix it? Waiting for your call."
- "Do both, or only the test isolation?"
- "👉 Next: writing the failing test A2." — and the turn ends right there.
- "Say the word and I'll run it."
- "Shorten MEMORY.md or leave it as is?"
Each one pulls you back to the terminal for a one-word "yes". Step away, and that is an hour of nothing.
This Stop hook for [Claude Code](https://code.claude.com) answers for you. When the next step is obvious, reversible and in reach, the agent gets the turn back and does it. You get controlled autonomy: the agent keeps going on its own, and you spend the time on strategy — or on your own things.
TL;DR
- What — at every turn end an LLM judge reads the final message. An excuse gets the turn back. It is not a code reviewer and never judges whether the work is right.
- Where it still stops — irreversible steps, real forks, anything only a human can do.
- Cost — about $0.89 per 1 000 judged turns in 16 days of real use.
- Safe by default — one return per chain; any judge failure lets the turn end normally. Every forced return rings a macOS notification, so you always know when the judge acted.
- Setup — mine: Claude Opus as the agent,
deepseek-flashas the judge. Adapt it to yours.
Quick start
Requires `bash`, `jq`, `curl`, `python3`, `perl`.
git clone https://github.com/mkdir-username/claude-handoff-control.git
cd claude-handoff-control && ./install.sh
export DEEPSEEK_API_KEY=sk-... # or HANDOFF_CTL_API_KEY; put it in your shell profile
Restart Claude Code afterwards. `./uninstall.sh` removes everything the installer added.
Whatinstall.sh changes
- Files — copies hooks and rubrics to `~/.claude/handoff-control/
Related Skills
Baton Pass
Pass the baton before your context runs out: a Stop-hook + skill that hands a long agent session off to a fres
Stop Rules
Holds your coding agent to your team's written coding rules. A stop hook for Claude Code, Codex, Cursor and te
Terminal Handoff
Persistent Claude Code session handoff and secure remote control from authorised devices, with automatic conti
Get Found Claude Agent
"Get-Found OS" Instead of 100 separate bots, build one orchestrator. It runs your 4 phases as its operating lo
Brotherme Handback
Take a decision and the work under it back into your own hands, with nothing lost
8sync
One Rust binary that runs Anthropic's Claude Code on autopilot, with rails: five levels of hands-off work (/do
Related Agents
Browser Check
Browser verification and screenshots (report pages, stats sites). Use instead of doing browser work in the mai
QA Handoff
Role: You are the QA + Handoff agent — the final gate. You run the deterministic packager over the exports, re
Router Planner
Turn an approved goal into an ordered plan grounded in the actual code. Read-only. Use before a change large e