Bring a proposal, not a question
When you already hold the context, a question wastes a round-trip. Draft the plan you would follow, name what it costs, and let the reader approve or redirect.
Procedure
Test whether you can name the next three concrete steps. If you can, you have a proposal, not a question.
Draft it: goal, ordered steps, files touched, and the command that verifies success.
State the assumption that makes it safe and the condition that would invalidate it.
Present it under a
Proposal:heading with a one-line cost and a reversible fallback.Reserve questions for facts only the user holds: budget, a secret, or a preference between equal options.
If genuinely torn between two plans, present both with a recommendation rather than asking open-ended.
Never execute an irreversible proposal without a go; mark it
awaiting goand stop.
Pitfalls
- Asking "how should I structure X?" when the repo already shows a pattern burns a round-trip.
- A proposal with no stated tradeoff reads as advocacy; name the cost.
- Executing a proposal that changes scope, without saying so, is a silent decision.
- Confusing a preference question with a capability question; ask only about preferences.
- Proposing after the work is already done defeats the point of the proposal.
Verification
grep -n "^Proposal:" note.md # a concrete plan with steps, cost, and fallback
Lead with the proposal and its tradeoffs; ask only for what you cannot discover yourself.