The Agent Skills Directory

[DATA_EXFILTRATION]: The _image_source function in scripts/render_report.py reads arbitrary local files from paths provided in the input JSON. Using Path(source).expanduser().read_bytes(), the script base64-encodes file contents into data URIs for embedding in the generated HTML. This allows a malicious input to read sensitive files such as SSH keys or environment variables and include them in the report artifact.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted JSON data to generate interactive HTML reports, creating a surface for indirect prompt injection. 1. Ingestion points: JSON input files processed by the render_report.py script. 2. Boundary markers: The script employs html.escape() for text fields but lacks explicit delimiters or instructions for the agent to ignore instructions within the data. 3. Capability inventory: The skill performs arbitrary file reads via the image state and writes the resulting HTML to the filesystem. 4. Sanitization: Implements schema validation and HTML entity encoding for output.
[COMMAND_EXECUTION]: The skill requires the execution of a Python script that performs file system operations and generates artifacts based on user-provided data.

render-html