jimubi-bigscreen
Pass
Audited by Gen Agent Trust Hub on Nov 3, 2026
Risk Level: SAFECREDENTIALS_UNSAFECOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [CREDENTIALS_UNSAFE]: Hardcoded signature secrets were identified in 'references/scripts/bi_utils.py' and 'references/scripts/dataset_ops.py'. These are vendor-specific tokens used to generate HMAC-MD5 signatures for authenticating API requests to the JeecgBoot backend.
- [COMMAND_EXECUTION]: The skill executes local shell commands using 'subprocess.run' in 'references/scripts/gen_all_comps.py' and 'references/scripts/proc_ops.py'. These operations are used for legitimate utility purposes, such as copying preview URLs to the system clipboard or orchestrating internal script calls.
- [INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection by ingesting data from external APIs, SQL databases, and user-provided files (Excel/CSV/JSON) without explicit sanitization or boundary markers. While no active injection content was found, the capability to read and process untrusted data into the agent's context is present.
- Ingestion points: API responses from vendor-specified mock servers and platform endpoints, SQL result sets from user-connected databases, and contents of uploaded data files.
- Boundary markers: Not present in instructions for data interpolation.
- Capability inventory: File system writes (via 'Write' tool), network operations (via 'urllib' and 'requests'), and command execution (via 'subprocess').
- Sanitization: No explicit data filtering or escaping logic is implemented in the provided scripts.
Audit Metadata