Mischief-Manager

SusLink integration spike — policy engine

Decision: For mischiefmanager / ethical ad filter, integrate SusLink via HTTP to the existing TiltCheck safety API (same pattern as apps/chrome-extension → /safety/suslink/scan) for v1.

Rationale: Keeps @tiltcheck/suslink and its dependency graph (event-router, ai-client) out of the thin mobile/WebView client; centralizes scanning, rate limits, and model updates on the backend.

Alternatives considered

Option	Pros	Cons
Workspace package @tiltcheck/suslink	On-device, no URL egress	Heavy deps; AI path needs keys; not ideal for minimal mobile binary
Heuristic-only extract (LinkScanner subset)	Small, offline	Misses AI moderation; duplicate maintenance
HTTP API (chosen)	Reuse production scanner; one place to update	Requires consent for URL egress; network latency

LinkScanResult → enforcement tier mapping

SusLink returns riskLevel: safe

suspicious

high

critical (plus reason, url, scannedAt).

Merge with taxonomy tiers (additive):

SusLink riskLevel	Suggested minimum tier from URL signal alone	Notes
safe	(no upgrade)	Base decision from taxonomy/creative rules only
suspicious	allow_log if user enabled reporting; else no change	Good for watchlists
high	blur	User can override per domain if false positive
critical	block	Aligns with scam/phish/impersonation

Final tier = max(strictness) among category rules, SusLink mapping, and user overrides.

Latency strategy

• On navigation or ad click-through resolution: call full scan when consent allows.
• For high-frequency prefetch: optional quickCheck (heuristic-only) if exposed by API; otherwise batch/debounce and cache by landingUrlHash TTL 5–15 min.

Privacy

• Disclose URL transmission in consent copy (pairs with privacy model issue).
• Do not send ad creative body to SusLink unless product scope expands; URL only for v1.

API contract (implementation note)

• Request: normalized HTTPS URL, optional userId for authenticated TiltCheck users.
• Response: LinkScanResult JSON; map fields into evidence-bundle-v1 suslink object.

Test cases (acceptance)

1. Typosquat gambling domain → at least high.
2. Known safe retailer → safe does not upgrade a benign CAT-03 creative to blur.
3. critical + user allowlist domain → allowlist wins (document product policy).

This site is open source. Improve this page.