Baton Pass — Development skill for Claude Code
Pass the baton before your context runs out: a Stop-hook + skill that hands a long agent session off to a fresh one to cut token cost.
How to install Baton Pass
This entry records only its repository, not the path inside it, so there is no
exact command to give. Open Rorogogogo/baton-pass and copy the folder into
~/.claude/skills/, or the file into ~/.claude/agents/.
What Baton Pass does
Pass the baton before your context runs out: a Stop-hook + skill that hands a long agent session off to a fresh one to cut token cost.
Alternatives in Development
- Hackingtool Plugin — Claude Code plugin: 183+ pentesting & OSINT tools from Z4nzu/hackingtool 1k ★
- Taskmaster — Stop hook for Claude Code that keeps the agent working until all plans and user requests are 100% complete 494 ★
- Benjamin Plus Skill — Benjamin-Plus: a measured token-efficiency skill for coding agents (−17.9% cost median, quality unchanged) 272 ★
README
🏃 baton-pass
**Preserve progress before context or usage limits interrupt the work.**
**English** | [简体中文](./README.zh-CN.md)
A local continuity tool with optional context and Claude 5-hour quota guards. At a safe boundary it reuses the same `baton-pass` skill to hand work to a *fresh* session. It is one dependency-free **Go** binary—no Python, Node, `jq`, network service, daemon, or Notchy dependency.
    
🧠 Why this exists
Every turn of an agent conversation re-sends the **entire context** to the model. As a session grows, two bad things happen:
- Cost climbs every turn. A conversation parked at 190K tokens pays for ~190K of context on each reply — forever, until you stop.
- Auto-compaction kicks in. Near the window limit your agent silently compacts the history into a lossy summary you didn't write and can't review.
`baton-pass` fixes both and can also hand off before Claude Code's real 5-hour usage window is exhausted. Context, quota, and manual handoff are independent:
Context guard ─┐
Quota guard ─┼─→ baton-pass skill → handoff document → batonresume
Manual request ┘
The runtime decides **when**; the existing skill decides **how**. No second quota-specific handoff system is introduced.
🏁 Think of a relay race: each session runs its leg, then passes the baton (the handoff doc) to a fresh runner instead of dragging the whole track behind it.
💰 How much does it save?
The recurring cost of a turn scales with context size. With prompt caching, a cached re-read costs roughly **10%*
Related Skills
Ship Off
Disable ship-sop auto-mode. Sets trigger.mode to "manual" so the agent-sop Stop hook (sop-stop-drift.sh, user-
Depictio Embl Admin Login
Auth as Depictio admin on an EMBL K8s deployment by injecting the on-pod long-lived token into the React /dash
Amt Logout
Sign out of AMT memory - revoke this device's hook token and clear the local cache.
Claude Usage Report
Claude Code stop hook: per-prompt usage report with token costs, context window, and rate limits
Takeoff
Create, resume, or clear a session hand-off baton (one per work line)
Cut Video
Claude Code skill: tighten long recordings — remove silences, ums, dead air. Preserves laughs and comedic paus
Related Agents
Context Auditor
Use proactively when the user wants to cut Claude Code token cost or asks "why is this session so expensive /
System Critic
Adversarial design-system review with fresh context — hunts premature abstraction, domain leaks into primitive
Part 11: Auto-Capture Hook (Stop Manually Writing Memory)
By Terp - Terp AI Labs --- The vault system from Part 9 is powerful, but there's a gap: you have to manually w