claude-handoff

A Claude skill that captures the entire current chat as a structured handoff document, ready to paste into Codex, Cursor, or any other IDE-based AI coding agent.

Two files out, identical content, different formats:

• .md for tools that render markdown
• .txt flat text for tools that prefer plain output

What makes it different

Most "export this chat" outputs are raw transcripts. This one is lossless distillation. Every detail survives, but reorganized so the next model can act immediately. Critically, every code block from the chat is labeled:

• [FINAL] the version currently in use
• [SUPERSEDED] replaced version, with the reason it was replaced
• [REJECTED] thrown out, with the reason
• [DRAFT] proposed but never finalized

The rejection reasons are mandatory. That is the whole point. Without them, the next model re-suggests the same wrong thing.

Triggers

Any of these will activate the skill mid-chat:

• /handoff
• /export-context
• /brief codex
• /give to cursor
• /dump
• "package this for codex"
• "export this chat"
• "make a handoff doc"
• "send this to cursor"
• "i wanna take this to another model"

It also triggers on looser phrasing when the intent is clearly to move the session somewhere else.

Output structure

# Session Handoff: {Project}

## TL;DR
## Project Context
## Session Goal
## Timeline
## Decisions
## Constraints, Gotchas, and Hard Rules
## Files Touched
## Code Artifacts          <-- every block, labeled by status
## Errors and Fixes
## Commands
## User Preferences and Style Rules
## Open Threads
## Next Steps
## Raw Notes

Install

Important: the skill file must stay named SKILL.md. Do not rename it to anything containing the word "claude" -- claude.ai blocks drag-and-drop installs of skills with "claude" in the filename.

Claude Code / Claude Desktop with file access

Drop SKILL.md into your skills directory:

mkdir -p ~/.claude/skills/handoff
curl -o ~/.claude/skills/handoff/SKILL.md https://raw.githubusercontent.com/iam25th1/claude-handoff/main/SKILL.md

Adjust the path if your setup uses a different skills location.

claude.ai (drag and drop)

1. Download SKILL.md from this repo
2. Keep the filename as SKILL.md (do not rename to claude-handoff.md or anything with "claude" in it -- the upload will be rejected)
3. Drag it into a conversation or project in claude.ai

The skill registers as handoff internally, which is what the trigger phrases hook into. The repo name and any wrapper folder name can be whatever you want, only the file itself has the naming restriction.

Usage

Mid-chat, when you want to wrap up and move to another tool:

/handoff

Claude will produce two files in the outputs directory:

handoff-{project-slug}-{YYYY-MM-DD}.md
handoff-{project-slug}-{YYYY-MM-DD}.txt

Then paste the contents into Codex, Cursor, or wherever you're continuing the work. A good first prompt for the receiving model:

Read this handoff doc, then continue from Next Step 1.

Behavior notes

• One-shot. The skill never asks clarifying questions. It infers the project name from the chat context.
• Verbatim discipline. File paths, error messages, command strings, and version numbers are preserved exactly. Only prose around them gets tightened.
• No em dashes anywhere in the output. Uses double hyphens or commas instead.
• Empty sections are marked _(none recorded in this session)_ rather than omitted, so the receiving model knows nothing was missed.

License

MIT. Do whatever.