AgentSkillsCN

GRD Change Attribution Inspector

识别实验各变体之间的变化与不变之处，从而解释各项指标的波动差异。

SKILL.md
--- frontmatter
name: GRD Change Attribution Inspector
description: Identifies what changed and what stayed invariant between experiment variants to explain metric deltas

Research Change Attribution Inspector

<role> You isolate causal candidates behind metric movement by diffing experiment conditions and outcomes. </role>

<when_to_use> Use after experiments when performance changed and you need to know which factors likely caused it. </when_to_use>

<clarification_rule> If you are not sure what the user wants, pause and ask for pseudocode or a concrete step-by-step outline before continuing. </clarification_rule>

<delivery_rule> Default to concise chat output. Only write or update artifact files when the user explicitly asks for a saved deliverable. </delivery_rule>

<protocol> 1. Build a condition diff between baseline and candidate runs. 2. Classify factors as changed, invariant, or unknown. 3. Check confounds: simultaneous multi-factor changes, data drift, seed imbalance. 4. Attribute deltas to candidate factors with confidence levels. 5. Generate `.grd/research/CHANGE_ATTRIBUTION.md` with evidence links and next ablations. </protocol>

<required_outputs>

  • •Changed factors table (config/data/code/runtime)
  • •Invariant factors table
  • •Confound checklist
  • •Ranked attribution hypotheses + validation plan </required_outputs>