The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to process external trading claims from potentially untrusted sources such as URLs, video content, or repository indicators to formulate strategy rules.
Ingestion points: The 'When the idea comes from outside' section in SKILL.md details how the strategist imports external claims into the agent's context.
Boundary markers: The process requires a mandatory 'Interview' phase and the creation of a RULES.md file to freeze definitions, which acts as a structured boundary between external data and code execution.
Capability inventory: The agent generates Python scripts (backtest.py) and proposes trades to other bots (e.g., Execution Trader).
Sanitization: The methodology requires manual review and explicit agreement on all parameters before proceeding to code.
[DYNAMIC_EXECUTION]: The workflow involves the agent dynamically generating a backtest.py script based on the user's strategy rules.
Evidence: SKILL.md instructs the strategist to write a 'readable, single-file backtest' and execute it against historical market data.
Context: This is a core function of the strategist role and is constrained to a specific workspace (/workspace/trading-desk/strategies/) using standard templates.
[SAFE]: The skill incorporates significant security best practices, including a strict 'Never' list that prohibits live order execution, mandates testnet for paper trading, and requires human approval for all trade proposals.

desk-strategy-lab