The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from external files like ANALYSIS_MODELING_REPORT.md and RESULTS_REPORT.md to generate paper content, creating a potential surface for indirect prompt injection.
Ingestion points: The agent is instructed to read various modeling and results reports to populate templates in SKILL.md.
Capability inventory: The skill utilizes Bash, Write, and Edit tools to perform its tasks.
Sanitization: Content from the external reports is interpolated into the paper sections without explicit sanitization or strict boundary markers.
[COMMAND_EXECUTION]: The skill provides a PowerShell script, scripts/package_supporting_materials.ps1, intended to be run by the agent to create submission archives.
The instructions in SKILL.md specify running the script with -ExecutionPolicy Bypass, which is a standard procedure for automating local tasks but is noted for its override of default security policies.
A review of the script confirms it performs legitimate operations such as file hashing, manifest generation, and 7-Zip archiving without any malicious network or system behavior.
[EXTERNAL_DOWNLOADS]: The skill references and acquiring templates from well-known competition sources (e.g., CUMCM, MCM). These references are for documentation and compliance purposes and do not involve untrusted remote code execution.

5writing