qe-learning-optimization
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill exposes a potential attack surface for indirect prompt injection by processing untrusted data during its pattern learning phase.
- Ingestion points: The skill ingests data from local
feedback/*.jsonfiles as part of thepatternLearner.learnprocess specified inSKILL.md. - Boundary markers: The instructions lack defined delimiters or clear directives for the agent to treat ingested feedback as separate, untrusted content.
- Capability inventory: The skill executes operations using the
aqeCLI and persists data to a vector database (agentdb), which could be influenced by malicious patterns. - Sanitization: There is no description of sanitization or validation routines to filter potential prompt injection attempts hidden within the ingested feedback files.
Audit Metadata