The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The installation instructions utilize uvx to download and execute code directly from a GitHub repository (monarchjuno/tradingcodex) and a PyPI package (tradingcodex). Executing unverified remote scripts allows for arbitrary code execution on the user's local system.
[EXTERNAL_DOWNLOADS]: The skill requires downloading multiple external components, including the main tradingcodex package and specific broker providers (e.g., tradingcodex-alpaca), which originate from unverified sources not recognized as trusted organizations.
[COMMAND_EXECUTION]: The skill implements a local CLI wrapper (tcx) that executes shell commands for system operations, database management, and starting a local Django server. These commands run within the user's shell environment.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data, such as news feeds and research markdown files, which are ingested into the context of multiple specialized agents with financial execution capabilities.
Ingestion points: Research files in the trading/research/ directory and data from external financial APIs.
Boundary markers: The skill relies on "approval gates" but lacks explicit system-level delimiters to prevent agents from following malicious instructions embedded within ingested research artifacts.
Capability inventory: The agents are authorized to draft financial order tickets and execute commands through the local tcx wrapper.
Sanitization: There is no evidence of automated sanitization or instruction-filtering for the markdown research content before it is read by the agents.

tradingcodex-investment-workflow