plan-input-validation
Pass
Audited by Gen Agent Trust Hub on Aug 25, 2026
Risk Level: SAFE
Full Analysis
- [SAFE]: The skill serves as a procedural guide for a security audit. It instructs the AI agent to map trust boundaries, identify gaps like missing schema validation or XSS vulnerabilities, and generate a remediation plan.
- [SAFE]: The skill enforces a 'plan-only' guardrail, ensuring that the AI agent does not modify code or configuration without explicit user approval of the proposed phases.
- [SAFE]: No network operations, remote script downloads, or unauthorized file access patterns were detected. The skill references standard security tools and libraries (e.g., Zod, DOMPurify, Stripe CLI) in a functional context.
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to process untrusted data (user-provided code and project files) to identify vulnerabilities.
- Ingestion points: The agent reads user code at various paths (e.g.,
api/webhooks/stripe.ts,Comment.tsx). - Boundary markers: The instructions explicitly state: 'Audit & plan only — no code changes until each phase is approved.'
- Capability inventory: The skill is limited to emitting a markdown report (
plan-input-validation.md) based on the analysis. - Sanitization: While the skill does not explicitly define sanitization for the inputs it audits, its limited capability to only generate a report minimizes the risk of malicious instructions in the analyzed code influencing the agent's actions beyond the audit scope.
Audit Metadata