What problem does it solve? Security review of pull requests in the LFX One application is easy to get wrong: reviewers either miss real authorization bypasses and token leaks or flood authors with speculative low-confidence findings. This Skill applies a diff-aware, high-confidence methodology tuned to the application's actual threat model so reviews catch real vulnerabilities without false-positive noise. ## Core Features & Use Cases - Diff-aware three-pass review: Context, comparative, and assessment passes trace attacker-controlled input to sensitive sinks, reporting only findings with confidence 8 or higher. - Durable threat anchors: Covers route classification in auth.middleware.ts, user-token vs M2M-token misuse, impersonation and effective-identity helpers, persona-based authorization, SSRF and redirect guards, XSS via DomSanitizer bypasses, PII logging, and secrets crossing the SSR boundary. - Per-fact data-exposure pass: Inventories every new or changed payload field, identifies its enforcement gate, and compares sibling read/write paths for parity gaps. - Use Case: A PR adds a new /public/api endpoint returning meeting data. The Skill checks the route's auth class, verifies private records are filtered before pagination, and confirms no user token or PII leaks to anonymous callers. ## Quick Start Review the security impact of the current pull request diff against the LFX One threat anchors and report only high-confidence findings.
