The Agent Skills Directory

[SAFE]: The skill's behavior is consistent with its stated purpose. It provides a structured way to interface with the Frevana API for hosting artifacts like HTML reports or documents.
[COMMAND_EXECUTION]: The skill invokes a bundled shell script (scripts/publish_file.sh) to perform its logic. This is a standard and transparent method for skill execution, utilizing common system tools such as curl and python3 to handle API interactions.
[DATA_EXFILTRATION]: Network operations are restricted to the official vendor domain (api.frevana.com) and target-specific pre-signed upload URLs. The data transmitted is limited to the user-specified file and associated metadata required for publishing.
[CREDENTIALS_UNSAFE]: The skill requires a FREVANA_TOKEN for authentication but handles it safely by reading it from the environment or command-line flags. It includes explicit instructions and logic to ensure credentials and temporary pre-signed URLs are not leaked in the output or stored locally.
[PROMPT_INJECTION]: The skill processes user-provided files to extract titles (Category 8: Indirect Prompt Injection surface).
Ingestion points: User-specified file content is read in scripts/publish_file.sh.
Boundary markers: None; the content is parsed directly for metadata.
Capability inventory: The skill has capabilities for network requests (curl) and file reading.
Sanitization: Content is processed using Python's html.parser and specific regex patterns to extract titles, which limits the risk of the agent executing embedded instructions from the file content.

frevana-publish