The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of a persistent knowledge base that influences agent behavior across different sessions. This workflow introduces an attack surface for indirect prompt injection.
Ingestion points: Data from research, competitive intelligence, and task investigations is saved into markdown files in the memory/ or wiki/ directories.
Boundary markers: The skill does not include instructions to use delimiters or ignore-markers for the content it stores, allowing potential malicious instructions in the data to persist.
Capability inventory: The skill writes, modifies, and deletes markdown files within the agent's workspace.
Sanitization: The instructions recommend excluding credentials but do not mention sanitizing the ingested text for prompt injection patterns before storage.

compile-knowledge