deep-research

Deep research over MFS-indexed sources

1. What this is (and what it replaces)

deep-searcher is an open-source framework that reasons over private data: it decomposes a question, iteratively searches a vector database, evaluates whether the evidence is sufficient, and synthesizes a cited report. Built before agentic coding tools existed, it had to hand-roll every piece itself — document loaders, a multi-provider LLM/embedding/vector-DB matrix, and a custom iterative-retrieval orchestration loop in Python.

None of that orchestration is needed anymore. MFS already does ingestion + hybrid search over many source types, and an agent's own reasoning loop already does "search, judge, follow up, repeat" natively once it has a search tool. This skill is that missing piece: not new retrieval code, just the strategy for running deep-searcher's decompose → search → evaluate → synthesize loop through mfs search / mfs cat.

This skill assumes mfs-find for the actual command mechanics (search modes, locators, --peek/--skim, index-status diagnosis). Read that skill for those details — this one only adds the multi-round strategy on top.

2. Precondition: sources must be indexed

Same as mfs-find: mfs status / mfs connector inspect <uri> first. If nothing relevant is indexed yet, redirect to mfs-ingest — don't run a research loop against an empty index.

3. The loop

 decompose            search rounds              evaluate           synthesize
┌───────────┐   ┌───────────────────────┐   ┌──────────────────┐   ┌───────────┐
│ 2-4 angles│ → │ mfs search per angle,  │ → │ enough coverage? │ → │  cited    │
│ on the Q  │   │ semantic + keyword     │   │ gaps → new angles│   │  report   │
└───────────┘   └───────────────────────┘   └──────┬───────────┘   └───────────┘
                        ▲                            │ not enough
                        └────────────────────────────┘ (max ~4 rounds)