The Agent Skills Directory

[PROMPT_INJECTION]: The skill documents a pattern for ingesting data from external files, creating an indirect prompt injection surface.\n- Ingestion points: The start_event_watcher function in SKILL.md reads lines from a JSONL file at a path determined by the application.\n- Boundary markers: While data is parsed as JSON, the example lacks boundary markers or instructions to ignore embedded instructions within the ingested string fields.\n- Capability inventory: The code performs file system reads (File::open, Seek) and emits events to the frontend context (app.emit).\n- Sanitization: The example uses JSON schema validation via serde_json but lacks content-level sanitization or filtering of string values to prevent injection.

tauri-file-watching-events