The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection attack surface due to its reliance on untrusted external data. * Ingestion points: The skill uses WebSearch and WebFetch tools in SKILL.md to retrieve data for analysis. * Boundary markers: There are no instructions defining delimiters or clear separation between retrieved content and agent instructions. * Capability inventory: The skill has access to Bash (for tool execution) and Write capabilities. * Sanitization: The instructions do not include requirements for sanitizing or validating retrieved data before processing.
[COMMAND_EXECUTION]: The skill uses Bash to invoke tools/decision_calculator.py. It directs the agent to interpolate user or web-derived inputs (such as win rates and gain amounts) into shell commands. This creates a potential command injection vector if the agent fails to ensure these inputs are strictly numeric.
[EXTERNAL_DOWNLOADS]: The skill performs network operations to retrieve content from arbitrary external domains using WebSearch and WebFetch during its 'Decision-Oriented Research' phase.

decision-algorithm