Troubleshoot Feishu and Agent execution

Start with what you can see in Feishu, then ask the operator to inspect the relevant component. A repeated message can enqueue more work, so check status before resending a request whose outcome is uncertain.

Symptom	First check	Next action
Bot never responds	Correct bot, real @mention in a group, bot access and setup	Try /amux in private chat; operator checks Bridge connection and app permissions
Group discussion gets no answer	Shared topics require a mention to trigger work	@mention the bot in the existing topic with a specific follow-up
Other members are denied	Owner-only/shared access policy	Owner checks the group's /amux settings; sharing the group is insufficient
A card appears, then launch fails	Selected expert, Runtime and native Agent login	Operator runs Agent diagnosis on the executing host
Work appears stuck	Pending permission card, active turn or queued input	Resolve the current request; use /status or context help before interrupting
Old topic refuses input	Ended binding or retired group conversation	Start a new group message/topic; do not expect an old topic to become a new conversation
Group stopped after experiment was disabled	Saved independent group settings are paused	Owner checks the experiment and applies any required restart
Memory unavailable	Saved vs running switch, old unbound conversation, Hub scope	Apply settings, start a new conversation; do not fall back to another user's memory
Schedule has no visible chat reply	Actual schedule/run/task records and result delivery arrangement	Read the Task result; creation does not guarantee a Feishu push

Operator diagnostics

On the affected installation:

amux system status
amux system doctor
amux system target list
amux system logs

system logs prints paths; inspect the indicated Hub/Runtime logs around the failure time. system doctor diagnoses without changing the installation. Match the Runtime to the Session/Task's target; a healthy local worker says nothing about a failing remote one.

Check the selected Profile on the executing host, replacing it as needed:

amux agent doctor builtin/codex

The normal doctor is static. --smoke deliberately runs a real model turn and may use account quota. For a missing executable or expired login, repair the native Agent in that Runtime's service environment before retrying.

Recover deliberately

If services are stopped, amux system start starts the configured installation. Before restarting, inspect the plan and finish active work:

amux system restart --plan

Then run amux system restart when ready. Do not use installer --clean as a routine repair; it deletes amux data. If required isolation is unavailable, fix helper readiness instead of weakening policy.

For a useful issue, include version, time, message mode, Profile, target, Session or Task ID, the exact error and relevant log excerpt with credentials removed. Do not assume a screenshot of a final answer proves which TaskRun produced it.