Prompt Governance — Development

Prompt Governance treats prompts as production infrastructure: it helps you version prompts, build prompt registries, run automated evals to catch regressions, and manage governed iteration with A/B testing and rollback. It produces registry designs, eval pipelines, A/B test setups, and governance policies.

engineering claude-code

📂 View skill on GitHub ↗ 🚀 How to use

✨ What it does

Designs file-based or database-backed prompt registries with versioning and audit trails.
Builds eval pipelines with golden datasets and multiple eval types (exact match, LLM-as-judge, etc.).
Defines pass thresholds and rollback playbooks for safe deployment.
Provides A/B testing methodology with stable assignment and success metrics.
Flags anti-patterns like hardcoded prompts and missing evals.
Outputs governance policies and prompt diff reviews.

🎯 When to use it

When prompts are hardcoded in code and changes require full deploys.
When prompt changes have caused silent quality regressions in production.
When you need to A/B test prompt variants with real users.
When you want to establish a versioned prompt registry with environment promotion.
When you need automated evals to gate prompt changes before deployment.

🚀 How to use

Trigger by asking to 'manage prompts in production', 'build a prompt registry', 'create an eval pipeline', or 'set up prompt A/B testing'. The skill will ask for current state, goals, and AI stack. Provide details about your prompt storage, team size, and LLM providers. Example prompts:

We have 50 prompts hardcoded in our Python app. Help us set up a prompt registry and eval pipeline.

We want to A/B test a new summarization prompt against the current one. What's the best approach?

📄 Output: Deliverables include registry file structures and schemas, eval pipeline designs with golden dataset templates, A/B test plans, and governance policy documents.

📦 Add this skill to Claude Code

# 1. Get the skills repo
git clone --depth 1 https://github.com/alirezarezvani/claude-skills /tmp/claude-skills

# 2. Copy this skill into your project (or ~/.claude/skills for all projects)
mkdir -p .claude/skills
cp -r /tmp/claude-skills/engineering/prompt-governance/skills/prompt-governance .claude/skills/prompt-governance

Skill source: engineering/prompt-governance/skills/prompt-governance/SKILL.md

⚠️ Good to know

This skill is not for writing or improving individual prompts (use senior-prompt-engineer) and assumes you have access to LLM APIs for evals and a team willing to adopt governance processes.

❓ FAQ

What is the minimum golden dataset size for production confidence?

At least 20 examples for basic coverage, but 100+ for production confidence.

How do I rollback a bad prompt in production?

Use a one-command rollback that promotes the previous version back to production status in the registry, then re-run evals to verify.

What eval type should I use for open-ended generation?

LLM-as-judge is recommended for open-ended generation, scoring quality on a 1-5 scale.

🤖 Overview, features, install steps and FAQ were generated from the project's SKILL.md on Sep 4, 2026. Always check the original source before running commands.