The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill possesses an attack surface for indirect prompt injection through its data ingestion and output capabilities. \n- Ingestion points: The skill instructions direct the agent to 'search the web' for market context and 'read project files' such as client briefs and RFP documents, which could contain malicious hidden instructions. \n- Boundary markers: The skill does not define specific delimiters or instructions to treat external data as untrusted content, increasing the risk of the agent obeying instructions found in external data. \n- Capability inventory: The skill has the ability to write files to the project directory (e.g., {project}-pitch-deck.md) and read existing workspace files. \n- Sanitization: There are no requirements or logic for sanitizing or filtering external content before it is processed by the model.

alterlab-pra-pitch-builder