agent-global-context-commit
Pass
Audited by Gen Agent Trust Hub on May 11, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [SAFE]: The skill establishes a dedicated local directory (~/.agent-global-context/) for persistent memory storage. It incorporates safety guidelines, such as excluding secrets and passwords from storage and requiring user confirmation before saving sensitive information.\n- [PROMPT_INJECTION]: The skill's architecture for storing session context (user input, command results, and project metadata) across sessions creates a surface for indirect prompt injection. Malicious data encountered in one session could be persisted and influence agent behavior in future sessions.\n
- Ingestion points: Processes user statements, command outputs, and configuration files as described in the workflow sections of SKILL.md.\n
- Boundary markers: While using a structured Markdown template for storage, the skill lacks explicit delimiters or 'ignore' instructions for the content being stored to prevent instructions in data from being executed.\n
- Capability inventory: The skill manages file read and write operations within its designated memory root directory.\n
- Sanitization: No explicit content sanitization or validation steps are defined before writing data to memory files.
Audit Metadata