AgentSkillsCN

quality-label

为信号手动添加TP/FP/UNSURE标签，并保留审计轨迹，以支持质量指标与模式挖掘。

SKILL.md
--- frontmatter
name: quality-label
description: Apply manual TP/FP/UNSURE labels to signals and persist an audit trail
  for quality metrics and pattern mining.
user-invocable: true
allowed-tools:
- Bash
- Read

quality-label

When to use

  • •A human reviewer wants to mark a signal as a false positive or true positive.
  • •You are curating a gold dataset for evaluation.

Inputs

  • •signal_id
  • •label (TP|FP|UNSURE)
  • •optional: reason, notes, labeled_by

Workflow

  1. •Inspect the signal context (raw_data + why it surfaced).
  2. •Label it: python -m ops.cli quality label <signal_id> <TP|FP|UNSURE> --reason "..." --notes "...".
  3. •Confirm: signal_quality_metrics has one row for the signal with label_source='manual'.

Outputs

  • •feedback_id and label summary.

Guardrails

  • •If you’re not confident, use UNSURE rather than guessing.
  • •Manual labels override inferred labels; that’s intentional.

References

  • •references/reference.md
  • •docs/QUALITY_OPS_ARCHITECTURE.md