Your product assumptions (or your agent's) — proven or killed.
Your coding agent makes UX assumptions every session. Sightspool holds each one to falsifiable evidence — stands watch on the PostHog you already have, and writes the verdict back into your CLAUDE.md. It only speaks when it has proof.
FREE & OPEN-SOURCE · CLAUDE CODE & CURSOR · NO SDK
The loop: introspect, stand watch, close it.
Introspect from the code
It reads what your agent ships and drafts the assumption behind each change — belief → falsifiable question → signal → threshold. Every A-### entry is one testable claim.
Stand watch on your own data
Each committed assumption is measured against the PostHog you already run. Below the sample it needs, it says “still gathering” — never a verdict it can't back.
Close the loop into CLAUDE.md
When the evidence lands, the verdict is written back into your agent's context file — so the next session builds on proof, not the same guess again.
No proof, no post
Every finding traces numerically to the evidence it cites. With no proof it stays silent — the default state is silence, not a plausible guess.
A human rules the hard calls
An agent can rule an assumption refuted, never verified. When the data can't settle it, it escalates to a person instead of bluffing a resolution.
Free, open-source, no SDK
The ux-assumptions plugin is Apache-2.0 and installs straight into Claude Code or Cursor. No snippet, nothing to instrument — the register is the product.
Questions, answered plainly.
The plugin's free. So is the first workspace.
Start on Builder Free and explore with sample data. Your 30-day Team Trial begins only when the first verified live source connects, so reading the product never burns the evaluation window.
Create a free workspace