The Agent Skills Directory

Indirect Prompt Injection (HIGH): The skill is designed to ingest and process untrusted external data (project codebase) using tools like Read, Glob, and Grep.
Ingestion points: Source files from projects like 'yourbench' or 'coordinatr' are read into the context of five different security agents.
Boundary markers: There are no explicit delimiters or system instructions defined to separate the audited code from the agent's operational instructions.
Capability inventory: The skill possesses Write permissions (to generate reports) and Task permissions (to orchestrate sub-agents), which could be exploited if the agent follows instructions found within the audited files.
Sanitization: No evidence of sanitization or filtering of the source code content is provided before it is processed by the LLM, allowing for potential 'adversarial code' to influence findings or actions.

security-audit