gal-tab/sir_albert/claude-handoff

Ported from mattpocock/skills (MIT License, see THIRD_PARTY_NOTICES.md), adapted for claude CLI permission modes.

Claude Handoff

Write a handoff summary of the current conversation so a fresh agent can continue the work. Instead of saving it, launch a background agent seeded with the summary as its prompt.

1. Pick the mode

Decide whether the next session should execute (write code, run commands, modify files) or only plan / discover (research, investigate, produce a plan or report without touching the repo):

• Ask, does the user's request (or $ARGUMENTS) describe implementing, fixing, building, or shipping something? → execution mode.
• Does it describe investigating, researching, scoping, or producing a plan/design without applying it? → plan/discovery mode.
• If it's genuinely ambiguous from the conversation, ask the user which mode they want before continuing — don't guess.

2. Ask which model to use — before launching

Before building the launch command, ask the user which model the background agent should run on (e.g. sonnet, opus, haiku, or a specific model id). Do not default silently: model choice affects both cost and capability for a session that will run unattended. Wait for their answer before proceeding to step 4.

3. Write the summary

Include a "suggested skills" section, which suggests skills that the agent should invoke.

Do not duplicate content already captured in other artifacts (PRDs, plans, ADRs, issues, commits, diffs). Reference them by path or URL instead.

Redact any sensitive information, such as API keys, passwords, or personally identifiable information — the summary becomes the agent's prompt.

If the user passed arguments, treat them as a description of what the next session will focus on and tailor the summary accordingly.

4. Launch

Always pass -n/--name with a descriptive name (e.g. --name "Fix login bug") — it sets the display name shown in the job list, session picker, and terminal title.

claude --bg --name "<descriptive name>" --model <model from step 2> --permission-mode <bypassPermissions|plan, from step 1> "<handoff summary>"

It starts in the current working directory and returns immediately; the user manages it with claude agents.