soia-pkm-alipan-curator
Pass
Audited by Gen Agent Trust Hub on Jul 21, 2026
Risk Level: SAFEPROMPT_INJECTIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface (Category 8). 1. Ingestion points: Drive scans (e.g., inventory/initial.scan.jsonl) contain untrusted filenames. 2. Boundary markers: Explicit delimiters are absent. 3. Capability inventory: The skill can move, rename, and delete files (scripts/apply_reclass.py). 4. Sanitization: Structured JSON is used, but it does not filter natural language instructions in filenames.
- [COMMAND_EXECUTION]: All external command calls use list-based arguments with subprocess.run, mitigating the risk of command injection.
- [EXTERNAL_DOWNLOADS]: The skill uses recognized vendor repositories (soia-team) for its components.
- [SAFE]: The tool includes a detailed 'Run Bundle' system that requires user confirmation for destructive tasks and provides an auditable history of changes.
Audit Metadata