Guidelines for modifying model logic, updating metrics, and validating changes with backtests.

Install

mkdir -p .claude/skills/model-change && curl -L -o skill.zip "https://agentskills.codes/api/skills/download/14228" && unzip -o skill.zip -d .claude/skills/model-change && rm skill.zip

Installs to .claude/skills/model-change

Activation

This is the description your AI agent reads to decide when to run this skill — the better it matches your request, the more reliably it fires.

Use for prediction logic, feature usage, backtests, or metrics changes.
71 chars · catalog description✓ has a “when” trigger
Intermediate

Key capabilities

  • Update prediction logic in models
  • Modify feature usage for models
  • Run backtests for new features
  • Change performance metrics for models

How it works

The skill guides changes to model components, emphasizing adherence to a read order for relevant documentation and strict validation through tests and backtests.

Inputs & outputs

You give it
Model prediction logic, feature definitions, or metric calculations
You get back
Updated model behavior, backtest results, or revised metrics

When to use model-change

  • Update model prediction logic
  • Run a backtest for a new feature
  • Modify performance metrics
  • Validate changes to model inputs

About this skill

Model Change

When To Use

Use for prediction logic, feature usage, backtests, or metrics changes.

Read Order

  1. .agent/CONTEXT.md
  2. docs/reference/models/index.md
  3. docs/reference/specifications/transformations.md
  4. scripts/generate_predictions.py
  5. scripts/run_backtest.py

Rules

  • Preserve the reference_date cutoff.
  • Keep changes band-agnostic unless the issue is truly source-specific.
  • Pair logic changes with tests or a narrow reproducible validation path.

Expected Validation

uv run pytest tests/test_models.py
uv run python scripts/run_backtest.py --band goose --model notebook --shows 10

Common Mistakes

  • Leaking future data into features
  • Changing metrics without updating docs
  • Forgetting to validate downstream scripts

When not to use it

  • When the change is not related to prediction logic, feature usage, backtests, or metrics
  • When the issue is not truly source-specific and requires band-agnostic changes
  • When there is no intention to validate changes with tests or a reproducible path

Limitations

  • Changes must preserve the `reference_date` cutoff.
  • Changes should be band-agnostic unless the issue is source-specific.
  • Logic changes must be paired with tests or a narrow reproducible validation path.

How it compares

This skill provides a structured process for model changes, including specific validation steps, which differs from an ad-hoc approach to modifying model components.

Compared to similar skills

model-change side by side with the closest alternatives in the catalog.

SkillInstallsUpdatedSafetyDifficulty
model-change (this skill)04moReviewIntermediate
llm-evaluation03moNo flagsIntermediate
Plate Evaluation06moReviewIntermediate
llm-evaluation62moNo flagsAdvanced

Try saying

Example prompts that trigger this skill in your AI assistant.

Search skills

Search the agent skills registry