session-research-designlisted

# Session Research & Design Conduct deep research and produce a design plan before task planning. Open with one sentence saying what you are about to do and what it will produce. ## Non-Negotiables 1. **Never guess scope.** If the user's ask is ambiguous, ask a clarifying question. Do not assume. 2. **Always propose alternatives.** Present 2-3 approaches with explicit trade-offs. Do not ship a single-option "recommendation" — the user cannot make an informed choice without alternatives. 3. **Research before planning, unless it is already done.** The research report is normally a prerequisite to the plan, and the two are never compressed into one step. Research may be skipped only when a recent research artifact for this topic exists, or the user states the problem is already understood — in which case the plan's Context section names what it relied on instead. **Framing (Step 1) is never skipped**; it is the cheapest gate in the chain. 4. **User approves section-by-section.** Do not dump the full plan and ask "looks good?". Present in sections and get feedback on each. 5. **Cite the codebase.** Every claim about "what exists" must reference files/line numbers. "The auth module handles tokens" is not acceptable without a citation. 6. **Everything attaches to a work item.** Design for a `SEQ` that already exists, or capture one first. Never write a dated design document that no work item names -- an artifact with no identity is one nobody can select, claim, verify, or super