The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes the verona-toolkit CLI tool. It uses bash arrays ("${CMD_ARGS[@]}") to pass arguments, which effectively prevents command injection vulnerabilities by ensuring user-provided parameters (like port numbers or network names) are treated as literal strings rather than executable shell code.
[EXTERNAL_DOWNLOADS]: The skill references official Xion network endpoints (e.g., oauth2.testnet.burnt.com). These are well-known vendor domains associated with the author, Burnt Labs, and are required for the primary function of performing OAuth2 authentication.
[CREDENTIALS_UNSAFE]: While the skill manages authentication, it does not hardcode any credentials. It describes standard security practices of storing tokens in the operating system's native encrypted keyring and using PKCE for the authorization flow. Local credential paths mentioned (e.g., ~/.verona-toolkit/) are used by the external CLI tool for persistent session management.
[PROMPT_INJECTION]: The instructions in SKILL.md provide clear guidance to the agent on preferring auth refresh over auth login for efficiency. These are operational instructions and do not attempt to bypass agent safety filters or override system constraints.
[DATA_EXFILTRATION]: No patterns of unauthorized data exfiltration were detected. Network operations are limited to the intended OAuth2 callback server on localhost and the official vendor authentication APIs.

verona-oauth2