The Agent Skills Directory

[COMMAND_EXECUTION]: The skill frequently executes a local script memory-manager.py to perform actions such as preparing content, verifying database connectivity, and logging stored items.
[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface due to its core functionality of ingesting and indexing external content for RAG (Retrieval-Augmented Generation) systems.
Ingestion points: The skill accepts external research URLs and imported reports as "Content to store" or "Source context" in the SKILL.md file.
Boundary markers: There are no instructions defining boundary markers or specific warnings for the agent to ignore embedded instructions within the ingested data.
Capability inventory: The skill relies on memory-manager.py for content handling and communicates with remote vector databases (Pinecone, Qdrant) through MCP tools.
Sanitization: Content is normalized and metadata is validated by the management script, but there is no evidence of filtering or escaping potential instructions hidden within the stored research or reports.

save-knowledge