The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses subprocess.run in its test suite (tests/test_audit_adjustments.py) and smoke test (validation/smoke.py) to execute its own analysis script for validation purposes. These calls use sys.executable and -B flags, targeting the local scripts/audit_adjustments.py file without passing unsanitized user input, making this a standard and safe development practice for CLI tools.
[DATA_EXPOSURE_AND_EXFILTRATION]: No evidence of credential harvesting or data exfiltration was found. The skill processes local CSV files and outputs results to either a user-defined JSON file path or standard output. Sensitive file paths (like .ssh or .env) are not accessed.
[UNVERIFIABLE_DEPENDENCIES]: The requirements.txt file is empty, and the project documentation explicitly states that the analyzer has no third-party dependencies for offline use. This minimizes supply chain risks.
[INDIRECT_PROMPT_INJECTION]: The skill ingests external data (CSV format) into a structured analysis engine. While the data is processed, it is treated as numeric or categorical input for deterministic calculations (price jumps and return mismatches). The skill does not pass this raw data back into an LLM context in a way that would allow for prompt injection; instead, it outputs a structured JSON report. The ingestion points are load_rows in scripts/audit_adjustments.py, and sanitization is performed via strict numeric parsing and regex validation for dates.

corporate-action-adjustment-auditor