learn
Pass
Audited by Gen Agent Trust Hub on Jul 1, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted conversation history and has the capability to write to the file system.
- Ingestion points: The skill analyzes the current conversation context to identify learnings in Phase 1.
- Boundary markers: There are no explicit delimiters or instructions to the model to ignore malicious commands embedded within the conversation text.
- Capability inventory: The skill is authorized to use 'Write' and 'Edit' tools, which allows it to modify documentation files.
- Sanitization: No content validation or sanitization is performed on the extracted insights before drafting.
- Mitigation: The 'Phase 4: User Approval (BLOCKING)' step acts as a mandatory human-in-the-loop check, preventing the skill from performing any persistent file operations without explicit user confirmation.
Audit Metadata