feat(review): wire a live AI-vision caller for confirmed visual regressions

[JSONbored]

Summary

• Part of the visual-capture convergence epic (Epic: extend automated visual-capture convergence (#577) to self-host + metagraphed #3607); wires the deliberately-deferred feat(review): advisory-only AI-vision analysis of before/after visual captures #4111 follow-up. src/review/visual/visual-findings.ts already shipped the gating/prompt/parse logic but explicitly deferred a live caller — this PR is that caller.
• New exported runVisualVisionForAdvisory in src/queue/processors.ts, called right after the visual-capture block (independent try/catch, so a vision failure can never affect the "Visual preview" section that block already rendered): resolves submitter reputation + the maintainer's BYOK provider key (mirrors runAiReviewForAdvisory's own resolution, same re-resolve-per-caller convention), runs evaluateVisualVisionGate, fetches the before/after PNGs for the routes it selects, and calls the real BYOK vision provider (Anthropic/OpenAI). A parsed finding becomes an advisory visual_regression_finding — strictly advisory, can never become a gate blocker (see the file's own header).
• New fetchShotContentBlock in src/review/visual/capture.ts — fetches an already-captured shot URL and returns it as an AiContentBlock, degrading to undefined on any failure so one broken image never sinks the call.
• Exported the pre-existing (already-tested) base64Encode helper from src/utils/crypto.ts for reuse rather than duplicating byte→base64 logic.
• Extracted the wiring into its own exported function (rather than inlining it in the giant maybePublishPrPublicSurface) specifically so it's directly unit-testable without driving the full webhook pipeline — see the new test/unit/visual-vision-wiring.test.ts.

Scope

• The PR title follows type(scope): short summary Conventional Commit format.
• This PR is focused and does not mix unrelated backend, UI, MCP, docs, dependency, and deploy changes.
• This follows CONTRIBUTING.md and does not reintroduce GitHub Pages, VitePress, site/, or CNAME.
• No single open issue this PR alone resolves. Related: feat(review): advisory-only AI-vision analysis of before/after visual captures #4111 (already closed — the module this wires) and the parent epic Epic: extend automated visual-capture convergence (#577) to self-host + metagraphed #3607 (visual-capture convergence), still open pending Enable + validate automated visual capture end-to-end for metagraphed #3611's broader end-to-end validation.

Validation

• git diff --check
• npm run actionlint
• npm run typecheck
• npm run test:coverage — 100% branch/line on every changed hunk in all 3 touched src/** files (verified via scoped --coverage.include runs before and after adding tests)
• npm run test:workers
• npm run build:mcp
• npm run test:mcp-pack
• npm run ui:openapi:check
• npm run ui:lint
• npm run ui:typecheck
• npm run ui:build
• npm audit --audit-level=moderate — 0 vulnerabilities
• New/changed behavior has unit tests for every branch: all 3 gate-decline reasons, the desktop-vs-mobile viewport selection, a missing-URL route (skipped), a failed image fetch (degrades to fewer images), all-images-failed (no provider call at all), a successful call with a published finding, an empty-findings response, an unparseable response, a non-2xx provider response, a thrown error from the BYOK lookup (caught, logged, never escapes), and a null author.
• Full local gate (npm run test:ci) green end-to-end: 11,920 tests passed, 0 failed.

Safety

• No secrets, wallet details, hotkeys, coldkeys, user PATs, private keys, raw trust scores, private rankings, or private maintainer evidence are exposed.
• Public GitHub text stays sanitized — the vision system prompt explicitly instructs the model never to mention rewards/payouts/wallets/hotkeys/coldkeys/trust scores, mirroring the existing text-review prompt's boundary.
• N/A — no auth/cookie/CORS/GitHub App/Cloudflare/session changes.
• N/A — no API/OpenAPI/MCP surface touched.
• N/A — no UI changes.
• N/A — no docs/changelog changes needed.

Notes

• Deliberately does NOT change the automated capture's viewport count/dimensions (still desktop + mobile, no theme forcing) — shot.ts's own #4109 comment already explicitly considered and rejected widening to a 3rd viewport by default (50% render-cost increase for every repo, every review), in favor of a future per-repo opt-in if ever needed. Out of scope here.

[superagent-security]

Superagent didn't find any vulnerabilities or security issues in this PR.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 93.72%. Comparing base (b8a1ba4) to head (8d7df44).
⚠️ Report is 5 commits behind head on main.
✅ All tests successful. No failed tests found.

Additional details and impacted files

@@           Coverage Diff           @@
##             main    #4141   +/-   ##
=======================================
  Coverage   93.71%   93.72%           
=======================================
  Files         385      385           
  Lines       36209    36240   +31     
  Branches    13266    13281   +15     
=======================================
+ Hits        33934    33965   +31     
  Misses       1618     1618           
  Partials      657      657           

Files with missing lines	Coverage Δ
src/queue/processors.ts	95.22% <100.00%> (+0.03%)	⬆️
src/review/visual/capture.ts	91.39% <100.00%> (+0.35%)	⬆️
src/utils/crypto.ts	100.00% <ø> (ø)

🚀 New features to boost your workflow:

• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[loopover-orb]

Important

🟪🟪🟪🟪🟪🟪🟪🟪🟪🟪🟪🟪

🔍 Gittensory is reviewing…

AI analysis is in progress. This comment will update when the review is complete.

_{🟩 Safe / merged · 🟦 Advisory · 🟨 Held for review · 🟥 Blocked / closed · 🟪 Reviewing}