AgentSkillsCN

learn

针对 CYNIC 的判断结果提供反馈,以优化未来的评估过程。当用户希望纠正某项判断、给出改进建议、标记为正确或错误,或协助 CYNIC 从失误中汲取经验时,可运用此技能。

SKILL.md
--- frontmatter
name: learn
description: Provide feedback on CYNIC judgments to improve future evaluations. Use when asked to correct a judgment, provide feedback, mark as correct/incorrect, or help CYNIC learn from mistakes.
user-invocable: true

/learn - CYNIC Feedback Loop

"φ learns from φ"

Quick Start

code
/learn <judgment_id> correct|incorrect [reason]

What It Does

Provides feedback to improve CYNIC's judgment accuracy:

  • Mark judgments as correct/incorrect/partial
  • Explain why the judgment was wrong
  • Suggest what the score should have been
  • Trigger weight calibration

Feedback Types

OutcomeWhen to Use
correctJudgment was accurate
incorrectJudgment was wrong
partialSome parts right, some wrong

Examples

Mark Correct

code
/learn jdg_abc123 correct

Mark Incorrect with Reason

code
/learn jdg_abc123 incorrect "missed security vulnerability"

Provide Score Correction

code
/learn jdg_abc123 incorrect --score 35 "should have been BARK not WAG"

Implementation

Use the brain_cynic_feedback MCP tool:

javascript
brain_cynic_feedback({
  judgmentId: "jdg_abc123",
  outcome: "correct|incorrect|partial",
  reason: "explanation of feedback",
  actualScore: 35  // What it should have been (0-100)
})

Learning System

CYNIC uses feedback to:

  1. Calibrate Weights: Adjust dimension weights
  2. Detect Biases: Find systematic errors
  3. Improve Accuracy: Learn from mistakes

Check learning state:

javascript
brain_learning({ action: "state" })

View detected biases:

javascript
brain_learning({ action: "biases" })

Trigger calibration:

javascript
brain_learning({ action: "calibrate" })

Impact

Feedback VolumeLearning Effect
10+ feedbacksInitial calibration
50+ feedbacksBias detection
100+ feedbacksRefined accuracy

See Also

  • /judge - Create judgments to learn from
  • /patterns - See learned patterns
  • /health - Check learning system status