← ClaudeAtlas

security-reviewlisted

Find vulnerabilities that cross a real trust boundary and produce a concrete result, with source evidence, a severity anchored to demonstrated impact, and the smallest fix that enforces the invariant. Use for security questions, a focused review of auth or permissions or tenancy, a vulnerability investigation, or an audit of a codebase before it ships.
Udhay-Adithya/skills · ★ 0 · Code & Development · score 70
Install: claude install-skill Udhay-Adithya/skills
# Security review Find the vulnerabilities that violate a real trust boundary, then hand over the source evidence, the priority and the smallest effective fix. Source-first and defensive. **A candidate without a concrete affected principal, resource or security outcome is not a finding.** That single rule removes most of what passes for security review. ## Modes **Guidance**, the default. A security question, a focused review of one area, triage of a specific finding. Use the relevant parts of this skill. Do not create an output directory or write audit artifacts. **Audit.** Only when an audit, a pen test, a comprehensive review, or a report artifact is explicitly asked for. Then work through every surface systematically and write the report. Where the request could mean either, ask one question before creating any files. ## Execution safety Source inspection is read-only, and that covers most of this work. Run anything the target controls, a build, a test, a fuzz case, a fixture, only where the environment can enforce no external network, a scratch-local home and temp directory, a read-only target, and explicit CPU, memory, file-size and wall-clock limits. Do not install dependencies or let a build fetch them. **Never probe a deployed endpoint, a shared service, production identity, another user's data, or a live control plane.** Do not test availability against anything shared. Do not spend paid quota. Stop at the minimum local effect that establishes the defect: