The Agent Skills Directory

[SAFE]: The skill implements a multi-step research loop (Generate, Backtest, Debate, Report) using authorized internal tools such as emerald_signals and graph_orchestrate. The workflow logic is consistent with the stated purpose of overnight quant research.
[COMMAND_EXECUTION]: The skill mentions executing backtests on a GPU node via data-science-mcp, but this is a routed tool capability rather than arbitrary shell access. There are no patterns suggesting the ability to execute unauthorized system commands.
[DATA_EXFILTRATION]: No network operations to external or untrusted domains were found. All data operations (writing to KG, generating reports) occur within the expected research environment.

overnight-research