sivtr-memory
Pass
Audited by Gen Agent Trust Hub on Jul 23, 2026
Risk Level: SAFECOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill provides numerous patterns for executing the
sivtrCLI tool (e.g.,sivtr search,sivtr filter,sivtr zoom) to access local work memory. These commands are integrated into the agent's workflow as a primary data retrieval mechanism. - [DATA_EXFILTRATION]: The skill interacts with sensitive local data including shell history and agent conversation logs. While intended for context retrieval, this data may contain credentials or proprietary information. The skill mitigates risk by instructing the agent to search narrowly and requiring user consent for sharing or remote mutations.
- [PROMPT_INJECTION]: The skill presents an indirect prompt injection surface by reading untrusted terminal output and previous AI responses. \n
- Ingestion points: Terminal logs and AI history records retrieved via
sivtrCLI commands (SKILL.md). \n - Boundary markers: No explicit delimiter or 'ignore' instructions are mandated in the skill's command patterns. \n
- Capability inventory: The agent typically possesses shell execution and file system modification capabilities. \n
- Sanitization: No content validation or filtering is specified for the retrieved data.
Audit Metadata