Handoff

Produce a continuation prompt for a fresh session or resume from one.

Output Rules

• Chat only (default mode). Write the prompt in a single fenced codeblock so it copy-pastes clean. Skip files entirely; write one only if the user explicitly asks. The Reusable Mode below is the sole file-writing exception.
• No length limit. Capture everything load-bearing from this session, no matter how long the prompt runs; completeness over brevity, a dropped decision costs more than a long prompt does. Hand off decisions and state, not file contents. Secret values never go in; name where a credential lives instead.
• If the user gives next-step instructions when invoking the skill, they go into next verbatim. Exception: an instruction like "finish this then write a /handoff" means finish the requested work first; that instruction is already done by the time you write the prompt, so it doesn't go into next.
• Default: the new session starts in the same cwd and picks up the same static context (system prompt, CLAUDE.md) as this one, so don't restate that content. If the handoff may be resumed from a different cwd, use absolute paths throughout.
• Never assume the new session shares memory from this one. Skills you invoked, roles you took on, constraints you set: all of it needs restating or re-invoking. Tag files the new session should read at the start with @file.
• No git repo: if the injected git status/git log output is empty (cwd isn't a repo), write branch, SHA, and status fields as N/A (no git repo) instead of guessing a branch name or SHA.

Continuation Prompt Template

# /handoff <title>

> resume: read `load first`, verify `state` against the repo, treat `decisions` as settled, start at `next` 1. repo state beats this prompt on conflict.