The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process unstructured external content (meeting transcripts, conversation logs, pasted documents) which is an untrusted data source. ● Ingestion points: The workflow in SKILL.md parses raw input and extracts entities from external text provided by the user. ● Capability inventory: The skill utilizes the 'write_note' tool to commit information to the knowledge graph and performs web research based on extracted entities. ● Boundary markers: While the instructions emphasize preserving content verbatim, they do not prescribe the use of clear delimiters or instructions for the AI to ignore embedded commands within the processed data. ● Sanitization: There is no mention of sanitizing or filtering the external content before it is processed or written to notes. ● Mitigation: A significant safety measure is included in Step 5, which requires the agent to present a proposal and obtain user approval before creating any new entities.

memory-ingest