deep-divelisted

# Deep Dive Multi-angle research via parallel sub-agents and multiple web sources, then synthesis with citations. Follows the lead-researcher → parallel sub-agents → synthesis pattern from Anthropic's [multi-agent research system](https://www.anthropic.com/engineering/multi-agent-research-system). ## Triage First Before spawning agents, decide: deep research or quick lookup? | Signal | Action | |--------|--------| | Multiple angles, comparisons, "state of X", trade-offs, "what are people doing" | Run the full process | | One-shot factual question, syntax lookup, "what does this return" | Answer directly with one search; do NOT use this skill | | Ambiguous | Ask: "Quick lookup or a multi-angle deep dive?" | When in doubt, ask. Burning 4 parallel agents on a question that needed one search wastes tokens and time. **Scout mode:** for an open-ended "map the option space" question, you may run a cheap first pass with `mode: "scout"` (breadth, single wave, no escalation) to discover the angles, then run a full `mode: "deep"` pass. Depth-vs-breadth was a wash in testing for well-scoped questions, so default to `deep`; use `scout` only for scoping. ## Process ### 1. Plan the angles as a DAG, then ASK Don't decompose once and fan out once. Real research questions have dependencies — one angle's answer shapes whether a second angle is even worth running. 1. List 3–5 distinct research angles. Default to 3; go to 5 only if the topic genuinely splits that many ways. 2. For each