mem-load
Pass
Audited by Gen Agent Trust Hub on Mar 30, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill instructions create a surface for indirect prompt injection by directing the agent to ingest and follow instructions from external documentation files.
- Ingestion points: Reads
knowledge/INDEX.md,rules.md, andhypotheses.md(SKILL.md). - Boundary markers: Absent; the agent is told to apply rules by default without being asked.
- Capability inventory: No dangerous tools (such as shell execution, network requests, or file writes) are invoked in the
SKILL.mdinstructions, though the agent maintains its platform-provided tool capabilities. - Sanitization: Absent; the skill does not define any mechanism to validate or filter the content loaded from the knowledge directory.
Audit Metadata