Nox-Lumen Mfgask_user
In one sentence
The AI isn’t flying solo—before ambiguous, irreversible, or high-risk steps, it stops and asks you via the ask_user tool.
Typical triggers
| Scenario | Why it must ask |
|---|---|
| Gerrit -2 (block merge) | Irreversible; impacts others |
| Delete KB docs or wipe a KB | Irreversible |
| Push internal code to an external repo | Compliance |
| Ambiguous intent (“delete that doc”—which one?) | Avoid mistakes |
| LLM confidence is low | Don’t hallucinate certainty |
Why not let the LLM decide alone
An intentional human-in-the-loop gate—unsafe irreversible acts require confirmation. Platform rule, not skippable by a SKILL alone.
Working with SKILLs
Each SKILL declares explicitly which steps invoke ask_user—for example code-review SKILL may require confirmation before negative scores.
The platform does not decide for the SKILL author—you set your own accountability boundary.