mindos-max-zh
Pass
Audited by Gen Agent Trust Hub on Aug 11, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill uses instructions to override default agent behavior, enforcing an aggressive memory strategy that triggers without user prompts.
- [EXTERNAL_DOWNLOADS]: The skill requires the installation of @geminilight/mindos from npm, which is a resource owned by the vendor.
- [COMMAND_EXECUTION]: The skill executes various filesystem commands via the mindos CLI to manage the knowledge base.
- [DATA_EXFILTRATION]: The tool supports a remote configuration mode that enables sending local data to a specified URL, which could facilitate data theft if an attacker-controlled endpoint is used.
- [PROMPT_INJECTION]: The skill has a surface for indirect prompt injection by reading from a shared knowledge base without sanitization.
- Ingestion points: mindos search and mindos file read in SKILL.md.
- Boundary markers: None identified.
- Capability inventory: File system write, delete, and rename operations via CLI.
- Sanitization: No content filtering or escaping is implemented.
Audit Metadata