| name | handoff-prompt |
| description | Generate a handoff prompt that transfers the mental model — the WHY and WHAT — so the follow-up AI can own the HOW entirely. TRIGGER when: user says "handoff prompt", "hand this off", "another context-clean AI", "pick up later", "delegate this task", or any variation requesting a transferable prompt for task continuation or delegation. |
Phase 0 — EVOLVE
Read evolution.md in this skill's folder. Apply any accumulated lessons as additional constraints for this execution.
Philosophy
Minimal instructions, clear intent. The follow-up AI is a peer, not an intern. Transfer the mental model, then get out of the way. Less is more — if the AI can infer it from the files, omit it. Point to paths; don't summarize contents. Self-check: "Am I telling the AI what to think, or giving it what it needs to think for itself?" If the former, cut.
What You Produce
A clipboard-ready outline, under 20–30 lines (the principle block at the end counts toward the budget).
# Handoff: [Short Task Title]
## Why This Matters
[User's actual motivation. What problem? What do they care about?]
## Current State
[File path pointers. No content summaries.]
## Success Criteria
- [Outcome-shaped bullets — observable conditions that will be true when the task is complete. Write *"L6 LOW-energy RMSE appears in the summary table"*, not *"add L6 split to eval script"*. Conditions, not actions. Keep 1–3 bullets.]
## Non-document facts (only if such state exists)
- [Anything non-inferable from the files: decisions the user made, overrides, rejected alternatives, off-repo responsibilities, declared off-limits, environment realities. Write as past-tense state, not imperatives — *"User has declared the server off-limits"*, not *"Do not touch the server"*. The context carries the why; the rule follows from it. Keep a bullet only if omitting it would cause the follow-up AI to re-open a closed question.]
## Principles of Paramount Importance
- **Zero Assumptions:** Never guess user intent. If multiple implementations exist or requirements are incomplete, **halt and use the `AskUserQuestion` tool** to gather explicit direction.
- **No Silent Assumptions:** Even when the task is requested, confirm the *method* if it wasn't specified.
Principles of Paramount Importance
These apply not only to the follow-up AI receiving the handoff, but also to you while drafting it:
- Zero Assumptions: Never guess user intent. If multiple implementations exist or requirements are incomplete, halt and use the
AskUserQuestion tool to gather explicit direction.
- No Silent Assumptions: Even when the task is requested, confirm the method if it wasn't specified.
