AgentSkillsCN

calibration

为高置信度输入设定决策阈值,或为低置信度病例预留保守的安全裕度。

SKILL.md
--- frontmatter
name: calibration
description: Applies decision thresholds for high-confidence inputs or enforces conservative safety margins for low-confidence cases
license: MIT
compatibility: opencode
metadata:
  audience: developers
  workflow: clinical-pipeline

What I do

I apply calibration thresholds that determine how conservatively the pipeline operates. If detection confidence is >= 0.65, I apply standard thresholds (0.65). If confidence < 0.65, I enforce conservative safety margins (0.50) to reduce false positives.

When to use me

Use this when:

  • Skin tone detection is complete and you need to set decision thresholds
  • You need to determine whether to use standard or safety calibration
  • You're balancing sensitivity vs specificity based on detection confidence

Key Concepts

  • Standard Calibration: Threshold = 0.65 for high-confidence inputs
  • Safety Calibration: Threshold = 0.50 for low-confidence inputs
  • is_low_confidence: Boolean flag routing to appropriate calibration
  • calibration_complete: State flag indicating calibration is done

Source Files

  • services/goap.ts: Calibration action definitions
  • types.ts: WorldState interface

Code Patterns

  • Standard-Calibration-Agent: { is_low_confidence: false } preconditions
  • Safety-Calibration-Agent: { is_low_confidence: true } preconditions
  • Both set calibration_complete: true effect

Operational Constraints

  • Safety calibration is MANDATORY for low-confidence cases
  • Threshold decisions affect all downstream fairness calculations
  • Always log which calibration mode was applied