babarot/dotfiles/handoff

Trust tier 3
  • community
Source check pending

Checking that the SKILL.md file is still in the repository.

Write a copy-pasteable prompt that hands this session over to another agent, with a summary and this session's ID so the next agent can read the full transcript when it needs detail.

Install

npx vskill@latest install babarot/dotfiles/handoff

SKILL.md

handoff

The user wants another agent (a new Claude Code or Codex session, maybe on another machine) to continue the work of this session. Write the prompt they will paste into it. Output it in the chat; do not write a file.

1. Find this session's ID and transcript

  • Claude Code: the ID is in $CLAUDE_CODE_SESSION_ID (this session: ${CLAUDE_SESSION_ID}). The transcript is ~/.claude/projects/<cwd with every non-alphanumeric character replaced by ->/<id>.jsonl; confirm the path with ls.
  • Codex: use $CODEX_THREAD_ID if it is set. Otherwise take the newest ~/.codex/sessions/YYYY/MM/DD/rollout-*.jsonl whose first line (session_meta) has this session's cwd; the ID is its payload.id.

If neither works, say so in the prompt instead of guessing an ID.

2. Write the prompt

A summary, not a transcript: enough for the next agent to start working, with the session ID as the way to dig deeper. Write it in the language the user has been using, addressed to the next agent. Include:

  • Goal: what the user wants in the end, and why if it matters.
  • Working directory, repository and branch (and worktree, if any).
  • Done so far: what changed and where (files, commits, PRs), and what was verified.
  • Decisions: what the user chose or ruled out, with the reason, so the next agent does not reopen them.
  • Next: the remaining steps, in order, starting with the very next one.
  • Open questions and pitfalls: unresolved points, dead ends already tried, anything the user must do by hand (e.g. commands that need sudo).
  • Source: the session ID, the agent it ran in, and the transcript path. Tell the next agent it may look up details there only when the summary is not enough: with the claude-recall MCP (recall_search, recall_export) if it has it, or by reading the transcript file directly.

Leave out secrets and tokens even if they appeared in the session.

3. Output

Put the whole prompt in one fenced code block (use four backticks if it contains a code block) so it can be copied as is.

Also copy the same prompt to the clipboard with pbcopy, through a quoted heredoc so nothing in it is expanded:

bash
pbcopy <<'HANDOFF'
<the prompt>
HANDOFF

Before the code block, add one line saying whether it is on the clipboard (if pbcopy failed, e.g. in a sandbox, say so). Nothing else is needed.

Security

  • Security scanPassed, 90 of 100Pattern checks on the skill's files.
  • Model reviewPassed

Quality evaluation

  • Task evaluationNot runNo evaluation yet. Run one in Skill Studio.

Source

Popularity

Repository stars
780
Repository forks
92
7-day trend score
33

Stars and forks are for the whole GitHub repository, not this skill alone. Not refreshed yet.

Works with

Any agent vskill installs to: 56 targets in all. These 39 have a page on this site.

  • Claude Code
  • Cursor
  • GitHub Copilot
  • Windsurf
  • Codex
  • Gemini CLI
  • Cline
  • Amp
  • Roo Code
  • Goose
  • Continue
  • Kilo Code
  • Junie
  • Trae
  • OpenHands
  • Qwen Code
  • AI Agent
  • Tabnine Chat
  • Sourcegraph Cody
  • Amazon Q
  • Firebase Genkit
  • Sweep
  • Devin
  • Phind
  • Mentat
  • Claude Desktop
  • Aider
  • GPT Pilot
  • Plandex
  • bolt.new
  • v0
  • Replit Agent
  • Anse
  • TypingMind
  • BetterChatGPT
  • LibreChat
  • Warp AI
  • Pieces
  • Fig AI
Embed badge
vskill VERIFIED badge
[![vskill VERIFIED](https://verified-skill.com/api/v1/skills/babarot/dotfiles/handoff/badge)](https://verified-skill.com/skills/babarot/dotfiles/handoff)

Something wrong with this skill or its page? Tell us.