Skip to main content
Quaero only knows what lifecycle calls tell it, so a missing agent means one of those calls isn’t arriving or is landing in a different workspace. Check in this order; most cases are solved by step 4.
1

Is the client signed in?

claude mcp list should show slaunt connected. For Codex, [mcp_servers.slaunt] should be in ~/.codex/config.toml and the browser login completed.
2

Is it the workspace you're looking at?

Have the agent call identity_context and compare workspace_name.
3

Are hooks installed and loaded?

Look for Slaunt entries in ~/.claude/settings.json or ~/.codex/hooks.json. Hooks load at startup, so restart after installing. Without hooks, the model has to call the lifecycle tools itself.
4

Did a run start?

Ask the agent which Slaunt tools it called. It needs a thread and a run (run_begin or the UserPromptSubmit hook).
5

Is it reporting progress?

About 45 seconds without run_step and the agent looks stale.
6

Codex only: is routing consistent?

Hook calls and tool calls must carry the same threadId; otherwise they land on different threads.
7

Is the workspace at its agent limit?

Extra agents are refused with Agent quota has been hit, not queued.
8

Was it archived?

Finished agents move to the archive.
A new row on every request means the client opens a new MCP session each time. Reuse Mcp-Session-Id. Several chats in one row is expected: chats from the same client merge into one agent, each as its own thread.