searchable-events
Pass
Audited by Gen Agent Trust Hub on Aug 19, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill documents an indexing mechanism for untrusted Nostr events that creates an indirect prompt injection surface.\n
- Ingestion points: Data enters the agent's context via the SearchableEvent interface across 129 event kinds, as detailed in references/searchable-kinds.md.\n
- Boundary markers: The documentation does not specify delimiters or boundary markers to prevent the agent from accidentally executing instructions contained within the indexed content.\n
- Capability inventory: The skill facilitates the agent's interaction with SQLite and Filesystem event stores that aggregate human-authored text from the public Nostr network.\n
- Sanitization: The documentation confirms that no markup stripping or content sanitization is performed on human-meaningful fields (titles, bodies, descriptions) before they are stored in the search index.\n
- Specific Risk Factors: Event kinds 30175 (PersonaEvent), 30176 (TeamEvent), and 30177 (ManagedAgentEvent) are explicitly designed to index
systemPromptandinstructionsfields, which are primary vectors for indirect prompt injection attacks.
Audit Metadata