The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external search results and saves the summary to a persistent memory file. This workflow introduces a surface for indirect prompt injection if the retrieved content contains malicious instructions designed to influence future agent actions when the memory is recalled. Ingestion points: Data enters from the inputs.user_message and the results of the multi-search-engine skill. Boundary markers: The composition steps do not use explicit delimiters or isolation instructions when interpolating retrieved content into the digest or memory storage. Capability inventory: The skill uses the memory_save tool to write data to the local file memory/morning-digest.md. Sanitization: Although xml_escape and truncate are applied to user inputs, these do not mitigate natural language instruction injection within the search engine results.

meta-scheduled-morning-digest