The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill accepts user-defined text via the --goal flag in CLI commands. This input is processed to categorize tasks into different model tiers or action lanes, which could be influenced by malicious content within the goal string.
Ingestion points: The goal argument in the npx jev route and npx jev turn commands in SKILL.md.
Boundary markers: Absent. The instructions do not define delimiters or specific boundary markers for the user-supplied goal text.
Capability inventory: The skill performs classification to route tasks between different LLM tiers (nano, fast, balanced, frontier, reasoning) and action levels (act, confirm, human).
Sanitization: Absent. There are no instructions for sanitizing or escaping the goal string before it is processed by the routing logic.
[COMMAND_EXECUTION]: The skill instructs the agent to utilize npx to execute the jev package from the npm registry.
Evidence: The documentation provides examples such as npx jev route --goal "..." and npx jev turn --goal "..." to perform routing and input guarding.

jev-route