The original title is about adding a security gate to SKILLmama's evaluation system. Let me rewrite this for a mobile feed.

Short summary

SKILLmama added security gates (v1.1–1.2) to evaluate libraries and agent skills differently, using hard rules to block dangerous candidates and soft SQP flags to warn without rejection. Agent skills face stricter gates because they ship instructions the AI interprets directly, unlike libraries that require code invocation. Security findings appear inline on result cards so users see trust signals alongside scores at the moment of decision.

• •Different threat models: libraries need code invocation to cause harm; agent skills give the AI instructions to execute immediately
• •Two-gate approach: hard rules block CVEs and credential theft; SQP soft flags (SQP-1 overly broad triggers, SQP-2 silent network calls, SQP-3 hardcoded locale) surface warnings
• •Security context appears inline on cards, not in separate reports, so it shapes the selection decision rather than being scrolled past

Generated with AI, which can make mistakes.