The Agent Skills Directory

Indirect Prompt Injection Surface: The skill is designed to ingest and interpret agent trajectories, which include user prompts and tool outputs from previous sessions. This creates a surface where instructions embedded within that data could potentially influence the analyzing agent's behavior. Ingestion points: Markdown files containing test trajectories (agent-metadata-*.md). Boundary markers: The instructions do not specify the use of clear delimiters or instructions to ignore embedded commands within the artifacts. Capability inventory: The agent is capable of executing shell commands via npm. Sanitization: No specific sanitization or filtering of the artifact content is described.
Command Line Execution: The skill involves running npm commands with parameters provided by the user. While this is a standard practice for development-oriented tools, it represents a point where local system interaction occurs based on external input.
External Resource Retrieval: The artifact collection process involves downloading data from external test environments. Users should ensure that the sources of these artifacts are verified and secure.

analyze-comparison-tests