The Agent Skills Directory

[SAFE]: The skill defines a highly specialized role with robust operational boundaries. It explicitly defaults to historical simulation and paper trading, with mandatory controls that prevent the agent from submitting or modifying live orders without verified authorization.
[COMMAND_EXECUTION]: Includes a Python validation tool (scripts/check_execution_orders.py) designed to audit CSV files for safety properties like participation limits, duplicate IDs, and tick-size compliance before the agent processes them.
[INDIRECT_PROMPT_INJECTION]: The skill processes external data such as broker logs and CSV order lists, which is a potential surface for indirect prompt injection. This is mitigated by the following:
Ingestion points: External CSV order files and broker execution logs (referenced in SKILL.md).
Boundary markers: Explicit instructions to treat incoming targets and risk limits as 'frozen inputs' and to use conservative fill assumptions.
Capability inventory: Analysis, simulation, and TCA reporting; the skill is constrained from performing live network-based trade execution without external approval.
Sanitization: The provided scripts/check_execution_orders.py script validates the structure and content of input CSVs, checking for common data errors and limit breaches.

quant-execution-microstructure