The Agent Skills Directory

[PROMPT_INJECTION]: The skill processes untrusted data from external documents and URLs, creating an indirect prompt injection surface where malicious instructions within a document could attempt to influence the agent's behavior.
Ingestion points: Content is ingested from local files or remote URLs via the witan read command as defined in SKILL.md.
Boundary markers: The tool outputs line-numbered text which provides basic structure, but it does not implement explicit delimiters or instruction-ignore warnings to sequester untrusted content.
Capability inventory: The skill can read local file systems, access network URLs, and is part of a workflow that includes writing to spreadsheets via witan xlsx exec.
Sanitization: No evidence of content sanitization or instruction filtering is described in the tool's operation.
[EXTERNAL_DOWNLOADS]: The skill is designed to fetch and process document content from remote URLs provided to the witan read command.
[COMMAND_EXECUTION]: The skill relies on executing the witan command-line utility, passing user-provided file paths or URLs as arguments to the subprocess for document conversion.

read-source