The Agent Skills Directory

Indirect Prompt Injection Surface: The skill's primary function is to ingest and analyze files from a repository (via targetPath). This introduces a potential surface for indirect prompt injection, where malicious instructions embedded in the source code or documentation being scanned could attempt to influence the agent's analysis or reporting.
Ingestion Points: Untrusted data enters the agent context through the files located at the specified targetPath and via the userContext field.
Boundary Markers: The instructions explicitly label the analysis data as "untrusted," which serves as an internal guideline for the agent's behavior, though it does not provide a technical sandbox.
Capability Inventory: The skill leverages the codex-security tool suite for coordination, discovery, and worker management. These tools perform the actual analysis and write the results to a report.md file.
Sanitization: The skill mitigates risk by ensuring analysis workers remain offline and use an immutable context snapshot captured at the start of the process, preventing runtime changes from affecting the worker's behavior.

deep-security-scan