What problem does it solve? Security reviews often devolve into generic OWASP checklists that miss real, exploitable issues in the actual code. This Skill structures a review around trust boundaries—where external data enters, where it flows, and where it gets treated as trusted—so findings are concrete and exploitable rather than theoretical. ## Core Features & Use Cases - Trust-Boundary-First Method: Enumerates all external inputs (HTTP requests, files, environment variables, database rows, third-party API responses, and model outputs) and traces them to dangerous sinks like shell commands, file paths, SQL, HTML templates, regex, deserialization, and redirects. - Credential and Permission Auditing: Checks for secrets leaking into logs, error messages, URLs, and diagnostic exports, and verifies authorization is enforced at the execution layer with default-deny and horizontal access control. - Sandbox vs. Access Control Clarity: Distinguishes language-level capability control from true isolation, which matters for agent systems running untrusted code. - Use Case: Before merging a pull request that adds a file-upload endpoint, run this review to trace the uploaded filename into shell commands and path joins, catching command injection and path traversal before deployment. ## Quick Start Ask the AI to perform a security review of the current codebase using the trust-boundary method and report findings ranked by exploitability.
