audit-integrity
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The 'Self-Learning System' defined in
references/self-learning-system.mdinstructs the agent to read context from files within the.github/SecurityLessonsand.github/SecurityMemoriesdirectories and apply that guidance before beginning any analysis. This mechanism allows content from the project filesystem to influence the agent's core methodology. - Ingestion points: The agent is directed to check for and load guidance from local filesystem directories (
.github/SecurityLessons,.github/SecurityMemories) at the start of every scan. - Boundary markers: The instructions lack specific delimiters or negative constraints to prevent the agent from inadvertently following malicious or contradictory directives embedded in those memory files.
- Capability inventory: The skill is intended for use by AppSec agents which typically have extensive capabilities including file system access, tool execution, and reporting, making the impact of an instruction override significant.
- Sanitization: There are no specified processes for sanitizing or validating the content of 'lessons' or 'memories' before they are ingested into the agent's active context.
Audit Metadata