longbkit/clisbot/clisbot-handoff

Handoff Skill

Transfer the current task — context, decisions, failed attempts, constraints — to a fresh agent. The receiving agent starts with zero context, so the handoff prompt must be a self-contained briefing.

User's arguments: $ARGUMENTS

Prerequisites

Read the clisbot skill. Call list_profiles before choosing the receiving agent. Do not create it until you have read the configured profiles and their notes.

Parsing arguments

1. Agent profile — explicit profile name first; otherwise choose the profile whose notes best match the work. Materialize it into create_agent as described by the clisbot skill. If no profile fits, use Clisbot's provider discovery fallback.
2. Isolation — "in a worktree" / "worktree" → create a workspace with isolation: "worktree", using a short branch name derived from the task.
3. Task description — anything else the user said.

The handoff prompt

The receiving agent has zero context. Include:

## Task
[Imperative description.]
 
## Context
[Why this task exists, required context.]
 
## Relevant files
- `path/to/file.ts` — [what it is and why it matters]
 
## Current state
[What's done, what works, what doesn't.]
 
## What was tried
- [Approach] — [why it failed or was abandoned]
 
## Decisions
- [Decision — rationale]
 
## Acceptance criteria
- [ ] [Criterion]
 
## Constraints
- [Must-not / must-preserve]

Preserve task semantics. Investigate-only → "DO NOT edit files." Fix → "implement the fix." Refactor → "refactor, not rewrite." Carry the user's exact intent.

Launch

Prepare the handoff in a dedicated workspace:

1. Select the current workspace or call create_workspace with the requested isolation.
2. Call create_agent with a [Handoff] <task> title, the briefing as initial prompt, and the selected workspaceId when explicit placement is needed.
3. Return the agent and workspace to the user, explaining that it remains in your subagent track until they detach it manually.

Do not encode independence as a create mode and do not invoke CLI or wire-level detach operations. Detach is a user gesture in the subagents track.

Do not wait or poll for the agent to finish.