The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests natural language input from the user and interpolates it into a prompt template, creating a surface for potential instruction injection through data.
Ingestion points: User input is provided via the task field in the qsh-form block of SKILL.md and interpreted by the agent during the workflow.
Boundary markers: The prompt_template uses {{#task}} delimiters to scope user input, which provides a basic structure but does not fully prevent adversarial instructions from being parsed.
Capability inventory: The agent has access to numerous financial data tools via the pandadata-api skill, including functions for fetching trade lists, financial reports, shareholder data, and market metrics.
Sanitization: The skill mitigates risks by requiring the agent to normalize criteria into a structured schema (atomic filters) and clarify material ambiguities with the user before executing any tool calls.

stock-screener