Falsifiable UX assumption tooling for founders who build with coding agents.
/plugin marketplace add sightspool/claude-plugins
/plugin install ux-assumptions@sightspool
Every shipped feature is a bet about users. Your onboarding bets people survive every step; your Pro gate bets it sits exactly where willingness to pay begins; your empty dashboard bets a first-time user knows what to do next. Products built fast with coding agents accumulate these bets faster than anyone writes them down.
/audit-assumptions reads your repo — routes, gates, empty states, error
paths, pricing logic, plus your README/CLAUDE.md and git history — and drafts
every untested belief it finds into ux-assumptions.md: a falsifiable
register where each entry carries the belief, the yes/no question that would
test it, the signal that would answer it, and a concrete threshold it must
clear.
- Zero account. Zero network. Zero telemetry. Your own agent does the analysis; nothing leaves your machine.
- Real provenance: every entry cites the file/line and artifact that implies it.
- Honest by construction: every drafted entry is
untested— the audit measures nothing, so it rules nothing. - Re-run safe: existing entries, ids, and verdicts are never rewritten.
The register format is frozen and versioned, so the file you get is portable — other tools (including Sightspool's paid register, which watches thresholds and rules verdicts) can read it, but the file is 100 % useful on its own.
MIT — see LICENSE.