---
name: "weights-biases-prompts-logger"
description: "Log, version, and evaluate LLM inputs, outputs, and dataset iterations with W&B Weave."
metadata:
  author: "Weights & Biases"
  version: "1.0.0"
  runtime_compatibility:
    - "codex"
    - "claude-code"
    - "cursor"
    - "windsurf"
triggers:
  - "log prompts with wandb weave"
  - "version prompt templates weave"
  - "track llm experiments weave"
permissions:
  fs_read: true
  fs_write: "workspace"
  shell_exec: false
  net_access: false
---

# Weights & Biases Weave LLM Prompt Logger

Integrates Weights & Biases Weave: decorating inference functions with @weave.op(), tracking prompt version iterations, comparing model outputs side-by-side, and publishing evaluation leaderboards.

## When to Invoke
Invoke this skill whenever you need to:
- log prompts with wandb weave
- version prompt templates weave
- track llm experiments weave

## Execution SOP & Rules
1. Initialize Weave project with weave.init('project-name') before running inference.
2. Decorate core pipeline functions with @weave.op() for automatic lineage and trace logging.
3. Compare prompt variations using Weave Evaluation datasets and scorer functions.
4. Inspect evaluation results and regression metrics on the hosted W&B dashboard.

## Verification Checklist
- Run local linting, syntax validation, and type-checks before reporting completion.
- Ensure all created or modified files adhere to workspace boundaries and avoid uncommitted secrets.
- Verify tests or build commands exit cleanly with return code 0.