co-tank-monitor
Pass
Audited by Gen Agent Trust Hub on Sep 23, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill package contains a metadata file with self-referential safety claims intended to bias the evaluation process.
- Ingestion points: The skill includes a simulated audit report (
co2-tank-monitor_audit_result_v2.json) and accepts numerical sensor data via CLI arguments inscripts/main.py. - Boundary markers: None. The skill does not implement delimiters or specific instructions to disregard instructions embedded within the included data files.
- Capability inventory: The functional logic in
scripts/main.pyis restricted to mathematical operations and formatted console output. It contains no network access, file system modification, or dynamic code execution capabilities. - Sanitization: CLI inputs are strictly typed as floats or integers by the
argparselibrary. However, the skill provides no validation or sanitization for the claims present in the included JSON audit file. - Evidence: The file
co2-tank-monitor_audit_result_v2.jsoncontains assertions such as"security": "PASS","deployable": true, and"grade": "Production Ready". According to the safety guidelines, these self-referential safety claims must be treated as untrusted data rather than authoritative conclusions.
Audit Metadata