The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process large amounts of untrusted text (academic drafts, papers, theses) provided by the user. This creates a surface for indirect prompt injection where malicious instructions embedded in the document could influence the agent's behavior.
Ingestion points: The create_workbook tool in SKILL.md takes a full markdown draft as the instruction parameter.
Boundary markers: No specific delimiters or instructions to ignore embedded commands are specified for the ingested draft.
Capability inventory: The skill uses a bash tool and workbook management tools (create_workbook, read_workbook).
Sanitization: There is no evidence of sanitization or filtering of the input text before it is passed to the analysis tools.
[EXTERNAL_DOWNLOADS]: The skill instructs the user to connect to an external Model Context Protocol (MCP) server at https://mcp.hedwigai.com/ and install plugins from the marketplace (un-seen/hedwigai-skills). These resources are hosted by the skill's author (un-seen).
[COMMAND_EXECUTION]: The skill utilizes a bash tool provided by the external MCP for processing files (e.g., doctext, pdftext) and querying scholarly records. While these are presented as utility functions for the skill's primary purpose, the presence of a shell environment allows for broader command execution.

reference-checker