The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes local Python scripts for data processing. scripts/compile_ae_rule_query.py implements secure identifier quoting and literal escaping to ensure that generated SQL queries are safe from injection attacks.
[DYNAMIC_EXECUTION]: The scripts/evaluate_rules.py script evaluates user-defined formulas using ast.parse. Security is enforced by a restricted node visitor that only allows a specific subset of mathematical operators and verified aliases, effectively preventing code injection through formulas.
[DATA_EXFILTRATION]: The skill includes an explicit masking mechanism for sensitive data. The mask_identifier function is used to redact IP addresses and account IDs in alerts and notifications, ensuring privacy compliance.
[INDIRECT_PROMPT_INJECTION]: The skill identifies a potential attack surface through external data ingestion but implements strong safeguards. 1. Ingestion points: External event data is ingested via CSV files in scripts/evaluate_rules.py. 2. Boundary markers: The instructions define clear gates and mandatory execution sequences, including project and schema resolution steps. 3. Capability inventory: Capabilities include read-only SQL query compilation and local rule evaluation. 4. Sanitization: Input is sanitized via strict SQL identifier/literal quoting and a hardened AST-based formula evaluator that permits only safe mathematical operations.

ae-risk-monitoring