The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data in the form of mathematical modeling competition problems and their associated data attachments.
Ingestion points: The skill instructions in SKILL.md explicitly direct the agent to read external "problem descriptions" and "attachments" to perform sub-problem decomposition and data understanding.
Boundary markers: There are no instructions for the agent to use specific delimiters (e.g., XML tags or block quotes) or guardrail prompts to distinguish between the untrusted problem data and its own system instructions.
Capability inventory: The skill has access to potentially high-impact tools including Bash(*), Write, and Edit, which could be abused if malicious instructions are embedded in the competition data.
Sanitization: No sanitization, validation, or filtering of the external content is prescribed in the workflow.

2analysis-modeling