The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill fetches and analyzes content from arbitrary external websites based on user-supplied URLs. This content is then processed by the agent to generate reports and recommendations, which allows a malicious website to inject instructions into the agent's context.
Ingestion points: Website content is retrieved using curl and custom scripts as described in the Workflow and Dimension sections of SKILL.md.
Boundary markers: No delimiters or safety instructions are used to separate the untrusted external content from the agent's system prompt or audit instructions.
Capability inventory: The skill utilizes Bash, Write, and Read tools, providing a significant surface for exploitation if an injection is successful.
Sanitization: The fetched content is not sanitized or filtered for malicious instructions before analysis.
[COMMAND_EXECUTION]: The skill instructions involve constructing shell commands (e.g., curl, bash, python3) using user-provided inputs such as $URL and $ARGUMENTS. If the agent does not strictly validate or sanitize these inputs, it could lead to command injection vulnerabilities.

website-health