The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data via the --content or --content-file arguments and transmits it to the Frevana API (ai-factory.frevana.com). This creates an attack surface where malicious instructions embedded in the report content could target the remote rendering service or influence the agent's behavior when processing the returned HTML.
Ingestion points: scripts/generate_report.sh reads content from user-supplied strings or files.
Boundary markers: The script uses basic JSON escaping but lacks explicit delimiters or instructions to the API to ignore potential commands embedded within the content payload.
Capability inventory: The skill utilizes curl for network communication and can write the resulting HTML to the local filesystem if requested via the --output parameter.
Sanitization: While json_escape is implemented to prevent structural JSON injection, no semantic sanitization of the content is performed.
[DYNAMIC_EXECUTION]: The generate_report.sh script employs a Python heredoc (`python3
<<'PY'`) to execute inline logic for parsing JSON responses. Although the Python code is static and used for data processing, it represents the generation and execution of script content at runtime.
[COMMAND_EXECUTION]: The skill uses bash to execute its logic and curl to perform network operations, which are the primary mechanisms for its functionality.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to https://ai-factory.frevana.com. While this is the expected behavior for interacting with the Frevana service, it constitutes an external network dependency.

frevana-gen-report