Turn messy UX research into a ranked, defensible backlog — with the receipts attached.
Throughline is a Claude Skill (Agent Skill) for UX research synthesis. Most research tools and AI summarizers stop at insight — themes, summaries, a tidy archive. Throughline carries through to the decision layer: it takes your raw qualitative research and produces a prioritized, evidence-traceable backlog of what to do next, where every item links back to the exact participant quote behind it.
It's the difference between a research archive and a decision you can defend in a stakeholder review.
Built for the solo designer-researcher — the designer who runs their own interviews, usability tests, and surveys and has no time to synthesize before the next sprint.
- What it does
- Why it's different
- Example output
- Install
- Usage
- What it is — and isn't
- FAQ
- Contributing
- License
Drop in your raw research — interview transcripts, notes, usability sessions, or survey exports — and Throughline runs the full synthesis-to-decision pipeline:
- Segments everything into atomic evidence units (verbatim quotes, attributed to participants).
- Clusters them into themes — the affinity wall you'd otherwise build by hand.
- Ranks those themes into a defensible backlog of actionable items (problems, opportunities, requirements), ordered by weight of evidence.
- Attaches the receipts — every item links to the verbatim quotes and participants that justify it.
Two rules keep the output trustworthy:
- Receipts on everything. Every claim traces to a real, verbatim quote — nothing invented or paraphrased into a "quote."
- Honesty over false confidence. Three or more participants is a Finding; one or two is a Hunch, labeled as such. It under-claims rather than over-claims, because your credibility depends on it.
| Most synthesis tools / AI summarizers | Throughline |
|---|---|
| Stop at themes and summaries | Carries through to a ranked, actionable backlog |
| Output an unordered list of insights | Prioritizes by weight of evidence |
| "Trust the AI" | Receipts attached — every item traces to a verbatim quote |
| Built for research-ops teams | Built solo-first, for a team of one |
From five interview snippets about a budgeting app's onboarding, Throughline produces:
## The decision — start here
Move bank-linking out of the first run. Four of five people quit there —
because they were asked to connect a bank before the app earned any trust.
## Ranked backlog
### 1. Defer bank-linking until after the user sees the app's value
Type: Requirement · Confidence: Finding · Evidence: 4 participants
Receipts:
- "it asked me to link my bank before I even knew what the app did." — P1
- "I bailed when it wanted my bank login on the first screen." — P4Ranked, labeled by confidence, and every line backed by a real quote.
Download throughline.skill and add it to Claude:
- Claude.ai / Claude apps: Settings → Capabilities → Skills → upload
throughline.skill. - Claude Code: add the
throughline/folder to your skills directory, or install from this repo.
The skill is built on the open Agent Skills standard, so it also works with other agents that support SKILL.md (Claude Code, Codex, Cursor, Gemini CLI, and more). Once installed it activates automatically when you ask Claude to turn research into a backlog, prioritize insights, or decide what to build.
Source:
throughline/SKILL.mdis the pipeline and rules;throughline/references/METHODOLOGY.mdhas the ranking heuristics and a full worked example.
In Claude, with the skill installed:
"Here are five interview transcripts from last week's research. Turn them into a ranked backlog of what we should do — and show me the evidence behind each item."
Paste or attach your transcripts and Throughline produces a throughline-synthesis.md: the ranked decision first, the affinity wall beneath it, and an honest note on what the research is missing.
Throughline is the decision layer: it works on whatever research fits in one conversation and turns it into a defensible backlog. It does not persist or accumulate research across sessions — there's no "ask my whole research history anything" memory here. That's a job for a full application, not a skill.
How do I install a Claude Skill? Download the .skill file and upload it in Claude's Settings → Skills, or drop the skill folder into your Claude Code skills directory.
Does it work with Claude Code / other agents? Yes — it's a standard Agent Skill (SKILL.md + references), compatible with Claude Code and other agents that support the open skills standard.
Is my research data private? Throughline runs inside your Claude session and works only on what you paste or attach. It writes nothing externally. Don't commit real research data to a repo — the included .gitignore excludes generated *-synthesis.md files.
Can it invent quotes? No — by design it quotes only what appears verbatim in your sources and drops any claim it can't back with real evidence.
Who is it for? Solo designers, design-leaning PMs, freelance researchers, and anyone who has raw qualitative research and needs to decide — and defend — what to do about it.
Issues, ideas, and PRs are welcome — see CONTRIBUTING.md. If Throughline helped you synthesize a study, a ⭐ helps others find it.
MIT — see LICENSE. Built by Ravi Palwe, who has spent nearly two decades designing the layer between AI logic and human trust.
Keywords: Claude Skill · Agent Skill · UX research · user research · research synthesis · qualitative analysis · affinity mapping · product backlog · research repository alternative · Anthropic Claude · design research · product management.