deep-research-queuelisted

# Deep Research Queue ## Purpose Defer research-worthy questions to the autonomous `deep_researcher` agent that runs nightly at 02:45 on the always-on Mac Mini. Each queued question becomes a multi-source synthesis report (≈800–1500 words, ≥3 citations) at $0/run via local LDR + Qwen3-14B + SearXNG, with the digest landing on the next morning's daily note. This skill is the **interactive write-side** of the deep-research pipeline; it does not run research itself. It only edits the queue file. The agent does the work overnight. ## When to Queue vs Answer in Session Use this decision tree before queuing: | Situation | Action | |---|---| | Single-fact lookup ("when did X ship?", "what's Y's price?") | **Answer in session** (or use WebSearch/WebFetch directly) | | Code or syntax question | **Answer in session** — wrong tool for LDR | | Multi-source synthesis with ≥3 citations needed | **Queue** | | Question with a deadline within an hour | **Answer in session** even if expensive — queue would miss the window | | Sean explicitly says "queue" / "add to research queue" / "research this later" | **Always queue** — defer to user intent | | Question requires reasoning over time-sensitive market data (prices, TVL, etc.) | **Don't queue** — the 02:45 fire would use stale-by-morning data; answer now or recommend a paid Perplexity/Claude.ai run | | Question is exploratory ("what should we think about X?") | **Queue** if the answer would benefit from breadth; otherwise discuss in sess