Wangxiaobao CLI Shared Rules

Problem

Wangxiaobao CLI business commands depend on login state, active project, tenant headers, and data permissions. If an agent runs commands directly or invokes blocking xiaobao-cli auth login, it may block polling, misread stderr, leak credentials into context, or mishandle NOT_AUTHENTICATED / NO_ACTIVE_PROJECT. This skill centralizes those cross-command constraints into a readable protocol so other Wangxiaobao skills establish the same invocation preconditions first.

How it works and key steps

• Non-blocking login: use --no-wait split-flow to obtain the OAuth device flow verification_link, paste it verbatim into the reply body, without URL encoding, rewriting, or code blocks; end the turn after sending and let the user authorize in the browser.
• Consume only stdout: parse success and failure results from stdout; treat stderr as human assistance only. Default toon optimizes context tokens; use --format json when strict machine-readable output is needed.
• Self-heal via hint: error objects include error, message, and hint. On NOT_AUTHENTICATED or NO_ACTIVE_PROJECT, follow the hint action rather than asking the user for guidance.
• Readable local state without exposure: tokens, active project, and pending auth are stored in local JSON files with 0600 permissions; agents must not print tokens and should use identity commands such as whoami for user info.
• Permissions and exceptions: business list commands are filtered by the current user’s authorization scope, kb * is isolated by active project, and admin * uses an independent whitelist; time values use yyyy-MM-dd HH:mm:ss.