The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the creation of adapters designed to ingest and sync content from external, untrusted sources (e.g., Email, Calendars, APIs) into ordinary nodes within the system. This architectural pattern exposes a surface where malicious instructions embedded in external data could attempt to influence the agent's behavior during processing.
Ingestion points: Data is ingested via the list, get, and get_changes operations within the adapter handler described in SKILL.md.
Boundary markers: The documentation refers to a "canonical contract" and uses mapping functions for normalization but does not explicitly define prompt boundary markers or instructions for the agent to ignore embedded commands in the synchronized data.
Capability inventory: The documented adapters have access to the raisin.http.fetch tool for network operations and are capable of performing file writes, updates, and command execution (via submit mode) within the mount system.
Sanitization: While the engine performs metadata enrichment and MIME type derivation, the skill documentation does not detail specific content sanitization or filtering logic for the data retrieved from providers.

raisindb-virtual-mount-adapters