uxaudit

Journey Compiler for the uxaudit pipeline. Translates each natural-language primary_journey from project-context.json into an executable JSON action script that capture_journey.mjs can replay against the running app. Reads project-context.json + curls the running app for selector hints, writes one journey-scripts/<id>.json per journey plus the mandatory _ftue.json. Single dispatch per iteration. Invoked from Phase 03 Step 9 of the orchestrator playbook.

L3-vision Judge for the uxaudit pipeline. Reads ONE captured screenshot plus a check-specific prompt.md (and optional evaluation brief) and writes a strict pass/fail/unverifiable verdict to result.json. Has Read and Write only — physically cannot drive a browser, run curl, or grep code. Invoked from Phase 04 Step 11 of the orchestrator playbook, one dispatch per L3 check, batched in parallel.

L4-journey Judge for the uxaudit pipeline. Reads the per-journey capture directory (screenshots + steps.json + evaluation brief) and writes a strict pass/fail/unverifiable verdict with four-axis journey_evaluation to result.json. Has Read, Write, Glob — no Bash, no WebFetch — physically cannot drive a browser or curl the app. Invoked from Phase 04 Step 11 of the orchestrator playbook, one dispatch per L4 journey, batched in parallel.

Project locator for the uxaudit pipeline. Maps a repository, enumerates dev-server candidates per sub-app (URL + start command + running state + auditability), identifies shared design-token packages, and flags out-of-scope native siblings. Writes project-locator.json. Pure file-reading + port-probing — no taste judgments. Invoked from Phase 01 Step 2 of the orchestrator playbook.

Proposer for the uxaudit pipeline. Runs after Reconciler, producing the cross-check synthesis. Compresses a long per-check failure list into 2-3 ranked, project-aware fix proposals anchored to root causes. The first (and only) subagent allowed to see BOTH the Scout's project-context.json AND the Judges' verdicts — the Judge's verdicts are already locked by the time the Proposer runs, so there is no rationalization risk. Writes improvement-proposal.json conforming to schemas/improvement-proposal.schema.json. Single dispatch per iteration. Invoked from Phase 05 Step 15 of the orchestrator playbook.

Reconciler for the uxaudit pipeline. Runs after every Judge has produced a verdict. The ONLY role allowed to see both the Scout's project-context.json AND the Judges' per-check result.json files. Flips catalog FAILS to PASS in strictly enumerated cases where the project's declared design system explicitly justifies what the catalog flagged. Writes reconciliation-summary.json and patches individual result.json files in place. Single dispatch per iteration. Invoked from Phase 05 Step 14 of the orchestrator playbook.

Project Scout for the uxaudit pipeline. Synthesizes an evidence-based hypothesis about a project's intent, primary user journeys, design system, and experience model. Reads README / spec / e2e tests / route definitions / harness state, then writes project-context.json. Does NOT make quality judgments. Invoked from Phase 01 Step 4 of the orchestrator playbook.

Run UX regression testing for a running app — combining static catalog checks (AI-slop, accessibility, Nielsen heuristics) with dynamic project-aware journey checks operated in pixel-only mode. Produces a unified editorial dashboard grouped into UX issues and UI risk signals, with cited sources, on-demand evidence, and improvement suggestions. Use this whenever you want to evaluate whether an app is understandable, polished, free of AI slop patterns, follows usability heuristics, AND completes its real user journeys end to end. Especially valuable after building a new app or iterating on UI changes.

1 hook across 1 event