human-review-queue

name	Human Review Queue
description	Flag the current agent decision or trace for human review in Label Studio. Use when uncertain, when a decision has significant consequences, or when operator validation is required before proceeding.
metadata	{"openclaw":{"compatibility":"skill-md"},"lv3":{"memory_refs":["memory:platform-context"]}}

Use this skill when the agent:

• Is uncertain about a decision and wants a human to validate it before acting
• Has taken a significant action (infra change, data mutation, access grant) that should be logged for review
• Has received a task that falls outside the governed tool boundary and needs operator sign-off
• Detects a conflict between two valid courses of action and cannot resolve it autonomously

How to flag a decision for review

Call submit-human-review-task with:

• prompt: The exact input or question the agent received
• response: What the agent decided or did (or would do)
• trace_id: The Langfuse trace ID for this interaction (if available)
• reason: Why human review is needed (uncertainty, consequence, policy edge)
• urgency: low | normal | high

The task will appear in the Label Studio annotation queue at https://annotate.example.com under the "Langfuse Trace Review" project. A human operator will label it as accept, reject, or needs_follow_up.

When NOT to use this skill

• Routine read-only queries (platform status, metrics, search)
• Tasks already covered by a governed approval contract (use Change Approval skill instead)
• Actions the agent is explicitly authorized to execute autonomously

Operator expectations

Tasks submitted here are reviewed within one business day. For urgent items (infra incidents, security findings), also send an ntfy alert via the notification tool.