The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The Python utility resources/payout.py operates entirely within the local execution environment. It uses the standard library only and contains no network-capable modules (such as requests or urllib). Financial data is processed via stdin or local JSON files, and no patterns of unauthorized data transmission were found.
[REMOTE_CODE_EXECUTION]: The skill does not download or execute remote scripts. It relies on a bundled Python script that is executed locally to process financial data. The instructions do not involve piping external content into shells or interpreters.
[DYNAMIC_EXECUTION]: The script includes functionality to load external benchmark and regression data via benchmarks_path or regressions_path. This is implemented using json.load, a safe deserialization method that does not allow for arbitrary code execution. Furthermore, the script performs strict type conversion to floats for all financial inputs.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted financial statement data (e.g., net income, capital expenditures). The attack surface is mitigated by the tool's architecture, which requires structured JSON and enforces strict numeric validation, preventing the agent from interpreting the data as instructions.

payout-policy-analysis