test-eyeballlisted

# Go eyeball — the QA loop on top of the capture Go and eyeball and FIX bugs using the debug browser, loop until you find no more, then hand back screenshots with example inputs. Do it as a loop, not a one-shot. **REUSE `/sk:ship-screenshot-changes` for the capture** — the debug-browser login (its Step 0), the port lane, figuring out what changed (Step 1), seeding the account into the state the change is seen in (Step 2), screenshotting each surface with realistic inputs (Step 3), the hand-back + open-in-Finder (Step 4), and the opt-in GitHub-native PR post (Step 5). Follow that skill for all of it; do NOT restate it here. This skill adds ONLY the QA a documentation pass skips: ## Drive each surface with EDGE inputs too, hunting bugs On top of the realistic pass (`/sk:ship-screenshot-changes` Step 3), also drive each changed surface with EDGE inputs — empty, very long, special/typographic chars (accents, em dash `—`, quotes, `& ( ) + . @`), boundary values, and the not-yet-configured states. These are where the real bugs hide (regex breaks, overflow, wrapping, NaN, XSS, wrong gating). Verify behavior with `evaluate_script` (read the DOM/computed styles/pill spans, not just eyeball), and check `list_console_messages({types:["error"]})` for feature-related errors (ignore pre-existing unrelated noise). Judge each surface against `~/.claude/references/user-journey-review.md` — the empty/loading/error states, a lock with no stated reason, a step with no way on. A surface that