iblai-api-agent-memory
Pass
Audited by Gen Agent Trust Hub on Sep 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill manages a memory system that creates an attack surface for indirect prompt injection by reusing untrusted conversation data in system prompts.
- Ingestion points: Conversation data from user messages is processed and stored as memories by the MemoryExtractionService referenced in references/concepts.md.
- Boundary markers: Memories are injected into system prompts using markdown headers like '## Relevant Context from Previous Conversations' as shown in references/concepts.md, but the system lacks explicit instructions or delimiters to isolate these memories from the rest of the prompt.
- Capability inventory: The skill provides full CRUD (Create, Read, Update, Delete) capabilities over memory stores via the api.iblai.app API documented in SKILL.md, allowing the agent to both read potentially poisoned data and write it back.
- Sanitization: No security-focused sanitization or filtering of memory content is evidenced in references/concepts.md or SKILL.md to prevent instructions embedded in memories from being executed during prompt injection.
Audit Metadata