uno-research

name	uno-research
description	Gathers context that does not exist yet: preliminary data analysis, user-study guides, SME and participant sourcing, Slack/analytics evidence sweeps, and codebase/design-system discovery. Instrument-first — the study guide is written before any conversation — and it hands cited findings to uno-synthesize rather than concluding. Use for fuzzy hunches ("tutors seem to churn — can we look into it?"), discovery questions ("do we already have a component for X", "where is", "how does"), sourcing knowledge-holders ("who actually knows how Y works?"), or evidence toward a go/no-go. Not for summarizing prior studies, ingesting pasted context, or drafting PRDs — that is uno-synthesize.
context	fork
user-invocable	true
argument-hint	["question / hunch / who-knows-X"]
allowed-tools	Read, Grep, Glob, Task, WebSearch, mcp__notion-plus__API-post-search, mcp__notion-plus__API-retrieve-page-markdown, mcp__notion-plus__API-get-block-children, mcp__notion-plus__API-query-data-source

uno-research — IDE face

Gather context that doesn't exist yet. The procedure lives in references/method.md — read it first. This file adds only IDE execution: which agents to summon, and what to load when.

When to use / when not

• Use for the fuzzy front end (no requirement yet, evidence needed), codebase/DS discovery, SME or participant sourcing, and preliminary data passes.
• Don't use when the context already exists — summarizing a prior study, ingesting pasted notes, or drafting a PRD routes to uno-synthesize. Edge cases fall to method.md's data rule: ingesting a prior analysis → synthesize; running new queries on raw data → research.

Agents it summons

researchers/explorer · researchers/source-miner · researchers/people-scout — defined in agents/ (see agents/README.md). Per the interaction contract, these are summoned by this skill, never by users. Study-guide writes go through writers/notion.

Workflow — method.md spine, IDE execution

1. Inventory first (method.md gate 1). Read docs/knowledge/INDEX.md; check the Research & notes DB for prior studies via mcp__notion-plus__*. Context already exists → route to uno-synthesize and stop.
2. Repo/DS questions → Task researchers/explorer. It owns the heavy reads and returns path:line-cited findings, never file dumps.
3. Org-evidence questions (Slack history, analytics, research DBs) → Task researchers/source-miner, naming the source types worth mining. Empty sweeps come back as findings.
4. People path — the gate binds before outreach, not before identification:
   1. Identification-only asks ("who knows X?", no conversation planned) may Task researchers/people-scout directly for ranked candidates.
   2. Before any intro draft is sent or conversation planned: draft the study guide per method.md; writers/notion writes it to the Research & notes DB. Read docs/conventions/notion.md before the write — convention surfaces only, never create select options. Notion unreachable? Draft the guide locally, flag it ⏳ pending Notion write, and say so — the gate is satisfied by the guide existing and being shown, never by pretending the write happened.
   3. Only then people-scout's intro draft goes out — sent by the human, who runs the conversation.
5. Data path → preliminary analysis scoped to the go/no-go question (method.md gate 2). No connected source for the data needed? Ask the user for an export — never fabricate a pull.
6. Compile the findings brief (method.md output shape) and hand to uno-synthesize. Do not conclude.

Tier-2 loads

Trigger	Load
always, on invocation	references/method.md
component / pattern / token discovery, before dispatching explorer	references/component-discovery.md
scoping an explorer dispatch — repo layout, key paths, commands	references/foundations-index.json
DS usage-pattern questions	references/patterns-index.json
prior-knowledge check	docs/knowledge/INDEX.md → the relevant lesson file
writing the study guide (via writers/notion)	docs/conventions/notion.md
pointing source-miner at Slack	docs/conventions/slack.md (channel map)

Quality bar

Rubric: docs/evals/rubrics/uno-research.md (study-guide-quality · analysis-scoping · sme-precision; hard gate: guide in Notion before any SME conversation). Golden scenarios: docs/evals/scenarios/uno-research.md — a change to this skill must keep all five passing.

Constraints

• Read-only, one exception: the study guide (and research notes), written via writers/notion to convention Notion surfaces only.
• Cite everything — a finding without a citation doesn't ship.
• Never contact people — people-scout sources and drafts; humans send.
• Never conclude — no takeaways, no go/no-go, no PRD. Hand the brief to uno-synthesize.
• Stay scoped to the question asked; expand into adjacent topics only when they change the answer.