The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the user or agent to execute shell commands such as jev-use install and jev-use doctor to configure the MCP server and verify its functionality.
[DATA_EXFILTRATION]: Data placed in the state parameter of the jev_judge and jev_gate tools is sent to external AI providers (such as TypeSafe, OpenRouter, or an AI Gateway) for processing. The skill specifically advises putting tool outputs and file excerpts into this field, which may contain sensitive information.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data (e.g., tool outputs, file excerpts) within the state field for external judgment.
Ingestion points: The state parameter in jev_judge and jev_gate tools (SKILL.md).
Boundary markers: No explicit markers are defined in the skill to prevent embedded instructions in the ingested state from influencing the judgment model.
Capability inventory: The skill provides tools for automated decision-making and gating, and requires shell command execution capabilities for setup.
Sanitization: No sanitization or validation of the ingested state content is mentioned.

jev-use