research-ops

Research Ops (thin pointer)

Operator wrapper around the repo's research stack. It does not duplicate those skills — it tells you which to reach for. The synthesis engine is deep-research; escalate to it whenever the answer needs multiple sources, citations, or confidence grading.

Route To

• deep-research — the orchestrator: multi-source synthesis, numbered citations, confidence grading, basic-memory persistence. Default for anything non-trivial.
• exa MCP (web_search_exa) — fast current-web discovery for a single quick fact.
• market-research — when the outcome must be a ranked recommendation / decision memo.
• lead-intelligence — when the real ask is people/company targeting, not generic research.
• knowledge-ops — when the result should be stored in durable context afterward.
• search-first — SEPARATE domain (research-before-coding: find existing libs/patterns, invokes the researcher agent). Use it, not this, for implementation research.

Discipline (applies whichever lane you pick)

• Don't answer current questions from stale memory when fresh search is cheap; don't spin up a heavyweight pass when local code/docs already hold the answer.
• Label every claim by evidence type: sourced fact · user-supplied context · inference · recommendation. Freshness-sensitive answers carry concrete dates.
• If the same lookup will recur, say so and recommend a monitor/workflow instead of repeating it.

The full loop (scope → preflight → dispatch → synthesize → reflect → self-eval → deliver/persist), untrusted-content rules, and output format live in deep-research — do not restate them here.