lite-plan

Interview the user relentlessly at an implementation level about every aspect of their idea, spec, or PRD until you reach a shared understanding of the build strategy, task order, code boundaries, risks, and verification approach.

Interview

Resolve every implementation decision, and keep the planned code minimal.

Walk down each branch of the decision tree for their implementation, resolving dependencies between decisions one-by-one.

Ask questions one at a time — exactly one question mark per message; a sub-question or "and also…" rider is the next message, sent after this answer lands — and wait for feedback before continuing. For each question, provide your recommended answer.

Proactively spawn Sonnet/GPT-5.6-Terra-Medium subagents to explore the codebase and investigate your own questions before resorting to asking the user: if a fact can be found in the codebase, look it up rather than asking. The decisions are the user's — put each one to them and wait for their answer.

A risk no check can retire is a decision: put it to the user during the interview, not into the artifact. The interview resolves open questions; the plan ships without any.

Readback

Once you've reached a shared understanding, read the plan back to the user before writing anything: one logical section per message, pausing after each so they can question, refute, or change it.