Skip to main content

Documentation Index

Fetch the complete documentation index at: https://orbit-docs.devotel.io/llms.txt

Use this file to discover all available pages before exploring further.

​
ADR-014: Playwright banned for routine UI verification — source audits preferred

Status: CURRENT Date: 2026-05-16

​
Context

Erol: “Playwright is time-loss”. UI tests were too slow for the verify-loop, flakier than reading the code, and agents were screenshotting pages that “looked right” but had a broken state machine underneath. Visual fidelity was passing while structural bugs (error fan-out, edge-case branches, state machine completeness) shipped.

​
Decision

No Playwright for routine UI verification of agent work. Verify via source-level mental-render audits: TanStack isError, status enum coverage, error-toast condition match, raw-token exposure, cross-view consistency, state-machine completeness, tab / active-state styling, timezone formatting.

​
Reason

Source audits catch the structural bugs Playwright misses and run in seconds, not minutes. Visual snapshot tests cannot detect the failure mode that actually matters here.

​
Consequences

  • Chrome DevTools MCP is allowed for one-off live probes after a deploy.
  • Playwright MCP is available but not part of the verify-loop.
  • The CLAUDE.md “Playwright for any UI claim” rule was demoted to optional 2026-05-13.
  • The scan_* memo family is the canonical artifact a source-audit produces.

​
Source memos

  • [[feedback_source_audits_not_playwright]]
  • scan_* memo family