The Agent Skills Directory

[SAFE]: The skill implements robust security guardrails by explicitly prohibiting the AI agent from requesting, seeing, or storing API keys. It utilizes a browser-based setup mechanism that keeps sensitive credentials entirely outside the agent's context.
[COMMAND_EXECUTION]: The skill relies on the local jev CLI tool to perform environment checks (jev doctor) and initiate the setup process (jev setup-key). These operations are local and consistent with the utility's intended purpose.
[INDIRECT_PROMPT_INJECTION]: The skill parses JSON output from local command execution. Ingestion points: Output from jev doctor and jev setup-key as described in SKILL.md. Boundary markers: None; the agent is instructed to expect specific JSON fields. Capability inventory: Local shell command execution and browser invocation. Sanitization: None; the agent parses output from a trusted local utility. The risk is evaluated as safe.
[PERSISTENCE]: The skill stores credentials in the OS secret store and local .env files to maintain state across sessions, which is standard behavior for setup and configuration tools.

jev-setup