The Agent Skills Directory

[SAFE]: The skill implements a 'Human-in-the-loop' safety architecture by enforcing mandatory checkpoints every three optimization rounds, preventing the AI from performing autonomous recursive changes without explicit user approval.
[SAFE]: Included Python scripts (scorecard.py, dashboard.py, evolution_log.py) are designed with a 'stdlib-first' approach, minimizing external dependencies and performing strictly local data processing, ranking, and journal logging tasks.
[COMMAND_EXECUTION]: The skill utilizes the Bash and Agent tools to execute local diagnostic scripts and spawn specialized sub-agents. These operations are limited to the strategy optimization workflow and are governed by a consensus-and-veto system where the 'Risk' and 'Backtest Integrity' seats can block unsafe proposals.
[SAFE]: The system includes explicit 'Backtest Integrity' and 'Simplicity & Robustness' guardrails that mathematically penalize strategy complexity and ensure performance gains hold up in out-of-sample data, mitigating common trading automation risks like curve-fitting.
[SAFE]: Strategy data and evolution logs are stored locally within the project directory structure (strategies/<name>/supercharge/), maintaining a clear and auditable history of all AI decisions without exfiltrating data to external servers.

fyers-supercharge