The Agent Skills Directory

[PROMPT_INJECTION]: No malicious override instructions or safety bypasses detected. The skill includes benign instructional constraints, such as an "anti-laziness" directive, intended solely to ensure complete analysis and report quality rather than to manipulate the agent's core safety logic.
[DATA_EXFILTRATION]: The skill targets official and commercial business data sources (e.g., gsxt.gov.cn, China Execution Information Open Network). There is no evidence of harvesting sensitive local environment variables, credentials, or private files. Data handling is limited to publicly available corporate information as per its professional use case.
[REMOTE_CODE_EXECUTION]: No remote script downloads or piped shell executions (e.g., curl | bash) are present. The included Python validation script (scripts/validate_monitor_report.py) is local, transparent, and relies only on the Python standard library.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from external business registries and news platforms. This is a standard risk for monitoring skills; however, the impact is mitigated by a multi-step workflow involving data cleaning and structured verification. No exploitable capabilities (like file system writes or network transmission of local data) are exposed to this ingestion surface.

loan-risk-monitor