uno-synthesize

name	uno-synthesize
description	Distills gathered context — Slack thread exports, interview transcripts, meeting notes, analytics pulls, docs — into a documented Findings & Takeaways doc and, on the designer's explicit go, a PRD with its paired uno-blueprint write. Use when the user says "synthesize this", "summarize the findings", "what are the takeaways", "what did people say", "what would we build", "draft the PRD", "write the update summary", or has collected raw context that needs distilling toward a build/no-build decision. Ingests existing context only — running new queries against raw data is uno-research.
user-invocable	true
argument-hint	["sources-to-distill | \"draft the PRD\""]
allowed-tools	Read, Grep, Glob, Write, Task, mcp__notion-plus__*

uno-synthesize — IDE face

The procedure lives in references/method.md (always load): ingest → findings toll gate → STOP at the designer's two gates → PRD → paired blueprint write. This file adds only IDE execution — sources, tools, and the agents this skill summons.

When to use / when NOT

• Use when context already exists — a pasted thread, transcript, notes, a prior analysis, referenced docs — and needs distilling: a scoped summary, a findings doc, a PRD, or a shipped-change update summary. Match depth to the ask (method § Scope): a quick "what did people say?" gets a summary with attributions, not the full pipeline.
• NOT for generating new evidence — new data queries, user studies, SME sourcing, codebase exploration → skills/uno-research. Not for building from an accepted PRD → skills/uno-prototype.
• Boundary test: does the context exist yet? For data: ingesting a prior analysis is synthesize; running new queries is research.

Sources — how context arrives here

Only Notion and Figma have MCP in this repo; everything else arrives as pasted text or an exported file/query result. Say so — never claim a connection that isn't there.

Source	Path in
Notion pages	mcp__notion-plus__* or paste
Figma frames	Figma MCP or link/paste
Slack · Zoom · Gmail · Box · Clarity · Metabase	paste / export only
Local files, repo, research briefs	Read / Grep / Glob

Execution — method steps → IDE actions

Method step	In the IDE
1 · Ingest	Read/Grep/Glob local material; pull referenced Notion pages via mcp__notion-plus__*
2 · Findings doc	summon writers/notion → Findings & Takeaways in the Research & notes DB
3 · STOP	present findings + recommendation; ask "enough context?" then "worth pursuing?"; wait for the designer
4 · PRD	draft inline for refinement; on approval, summon writers/notion → instantiate from the PRD template
5 · Paired write	on acceptance, summon writers/blueprint → stories/flows/screens to uno-blueprint in the same action as the PRD write; Roadmap card → Design Status: Ready for Design

No-go and not-sure outcomes at step 3 are deliverables too — see method § 3.

Agents it summons

writers/notion · writers/blueprint — defined in agents/ (see agents/README.md). Per the interaction contract they are summoned by this skill, never by users. writers/blueprint enforces the paired-write contract; writers/notion enforces convention surfaces and exact-match selects.

Tier-2 loads

When	Load
always	references/method.md
any Notion write	docs/conventions/notion.md
any blueprint write	docs/conventions/supabase.md
human-facing prose	docs/conventions/writing-style.md

Quality bar

docs/evals/rubrics/uno-synthesize.md scores every run — one hallucinated finding is an automatic fail; PRD + blueprint updating together is a hard gate. Golden scenarios: docs/evals/scenarios/uno-synthesize.md.

Constraints

• Distills, drafts, and recommends; never makes the go/no-go call and never implements code.
• Stays scoped to the provided context — expanding into adjacent evidence-gathering is uno-research's job.
• All external writes go through the writers; follow docs/conventions/notion.md (never invent select options).

Hand-offs

• PRD accepted (paired write done) → skills/uno-prototype (Flow 2 entry).
• Not worth pursuing → the documented sunset/postpone evidence closes the flow.
• Not sure / thin evidence → more gathering, or skills/uno-research for new evidence.
• Learning surfaced → skills/uno-maintain (knowledge capture).