omh-memory-new
Pass
Audited by Gen Agent Trust Hub on Nov 9, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill captures untrusted user input to be stored as durable project memory via the
omh_memorytool. This creates a surface for indirect prompt injection where a malicious user could attempt to store instructions that influence the agent's behavior in future sessions. - Ingestion points: User prompts and requests to "remember" facts, decisions, or preferences within SKILL.md.
- Boundary markers: The skill defines "evidence boundaries" and "not-evidence boundaries" to distinguish between instructions and data.
- Capability inventory: The agent has the capability to write to and correct a persistent memory store using the
omh_memorytool. - Sanitization: The instructions include proactive defenses such as summarizing input in the agent's "own words", converting relative dates to absolute, and explicitly refusing "prompt-injection-shaped instructions", secrets, and unsafe content.
- [COMMAND_EXECUTION]: The skill describes the use of the
omhcommand-line utility for status recording and memory operations. - Evidence: References to
omh runtime recordandomh_memorytool calls within the workflow definitions in SKILL.md.
Audit Metadata