The Agent Skills Directory

[SAFE]: The skill implements a robust auditing and governance framework for financial benchmark data. All functions are performed via local Python scripts that manipulate and validate structured data (YAML/JSON).
[REMOTE_CODE_EXECUTION]: No remote code execution vectors were found. The skill uses yaml.SafeLoader (via the _UniqueKeyLoader subclass in ag_common.py) to safely ingest data, protecting against unsafe deserialization attacks.
[DATA_EXFILTRATION]: No network-related imports (like requests or urllib) or command-line network tools (curl, wget) are present. Data processing is confined to the local filesystem for the purpose of auditing local benchmark bundles.
[COMMAND_EXECUTION]: The Python scripts do not invoke subprocesses or shell commands. They use standard library modules (pathlib, json, yaml, hashlib) for deterministic file manipulation and validation.
[PROMPT_INJECTION]: The skill's instructions focus on enforcing data integrity, immutability of historical results, and strict boundary rules between evaluation and governance layers. These instructions serve as safety constraints rather than attempts to bypass agent filters.
[CREDENTIALS_UNSAFE]: No hardcoded API keys or secrets were found. Example files use clearly identified dummy hashes (e.g., sha256:aaaaaaaa...) for demonstration purposes.

fin-kg-assure-govern-benchmark