The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process live browser text and accessibility data from external websites. This content is untrusted and could contain hidden instructions (indirect prompt injection) aimed at influencing the agent's decision-making process.
Ingestion points: Browser and desktop observation data gathered via host tools (SKILL.md).
Boundary markers: The instructions advise the agent to remove unrelated content and secrets, but do not provide specific technical delimiters or escaping mechanisms to prevent the LLM from obeying instructions embedded in the webpage text.
Capability inventory: The skill uses the jev-decide CLI to select actions, which are then executed by the host's native browser/system tools.
Sanitization: Relies on the agent manually filtering secrets from the observation state before processing.
[COMMAND_EXECUTION]: The skill requires the use of the jev-decide CLI tool to perform its primary function of evaluating next-action choices. It provides specific command patterns for validation (--dry-run) and live execution.
[EXTERNAL_DOWNLOADS]: The documentation references external resources for setup and workflow guidance, including the vendor's GitHub repository and established open-source projects in the browser-automation space.

jev-ui