Summary

• add native Feishu send_update_prompt support so gateway /update prompts use interactive Yes / No cards instead of plain-text fallback
• handle card callbacks with adapter-owned resolution, including duplicate/resolved prompts, unauthorized clicks, and callback scheduling failures
• persist update answers to Hermes's .update_response file and resolve the card inline
• add update prompt card behavior and resolution flow to the Feishu messaging docs

What changed

• added update prompt state tracking to the Feishu adapter
• added Feishu card builders for update prompt and resolved states
• added callback handling for hermes_update_prompt_action
• added safe scheduling failure handling in the shared loop submission path
• added authorization checks for interactive operators based on the existing Feishu allowlists/admin config
• added update prompt card behavior and resolution flow to the Feishu messaging docs

Tests

Ran:

• scripts/run_tests.sh tests/gateway/test_feishu_approval_buttons.py -q

Passing coverage includes:

• send_update_prompt sends an interactive card
• prompt state is stored after send
• not-connected send path returns a failure result
• Feishu send exception / timeout returns a failure result
• update prompt callback returns a resolved Yes card
• update prompt callback returns a resolved No card
• missing update_prompt_id is ignored safely
• already-resolved prompt callbacks return no card and do not reschedule work
• callback scheduling failure returns no card
• unauthorized operator clicks return no card and do not schedule work
• resolver writes .update_response
• resolver overwrites an existing .update_response file atomically
• unknown prompt IDs are dropped silently

Notes

• this keeps the existing Feishu approval flow pattern: interactive send, synchronous callback intercept, inline resolved card response, and async resolver side effects
• authorization reuses the existing FEISHU_ALLOWED_USERS and Feishu admin config; no new config is required