Getting Started
Welcome to Presient Labs. We optimize AI skills and prove the improvement with benchmarks.
What is skill optimization?
Submit a .md skill file, and we run it through our evaluation pipeline. You get back an improved version plus a report card showing exactly what changed and why it performs better.
Supported platforms
What you get
- →Optimized .md skill file
- →MCP server config (where applicable)
- →Ready-to-paste system prompt
- →Report card PDF with benchmark results
What Are AI Skills?
AI skills are the instructions that shape how AI models behave for specific tasks. They go by many names:
- →System prompts — instructions passed to the model before your message
- →CLAUDE.md files — project-level instructions that Claude Code reads at session start
- →.cursorrules / .windsurfrules — editor-specific behavior overrides for Cursor and Windsurf
- →Custom Instructions — ChatGPT’s equivalent, set in settings or Custom GPTs
- →Project Instructions — task-specific configuration in tools like Cursor, Windsurf, or custom agents
Why They Matter
These files control the quality of every AI interaction. A well-written skill produces better code reviews, more thorough debugging, more creative brainstorming. A mediocre one produces mediocre results — and most people never know the difference because they have no way to measure it.
The Problem
Most skills are written once, informally, and never revised based on evidence. The person who writes them has no way to know if their skill is performing at 60% or 90% of its potential. There's no benchmark, no A/B test, no feedback loop.
What Optimization Does
We take your existing skill and run it through a rigorous evaluation pipeline:
Benchmark the original
Measure pass rates across binary (pass/fail) criteria with 3 independent AI judges.
Mutate and test
Generate variations and evaluate each one the same way.
Keep what works
Only improvements that beat the original on hard metrics survive.
Deliver proof
You get the optimized skill plus a full report card showing exactly what improved and by how much.
The result
A measurably better skill, with evidence — not vibes.
How It Works
Your skill runs through our evaluation pipeline. We keep what works. We improve what doesn't.
Submit your skill
Upload your .md skill file through the dashboard. Works with any AI coding assistant format.
We benchmark and refine
Our pipeline evaluates your skill against real-world tasks and systematically improves clarity, specificity, and performance.
Receive your results
Download your optimized skill, system prompt, MCP config, and a full report card showing the benchmark improvements.
Evaluation Methodology
Our eval pipeline is designed to produce results you can trust. No vague scores — only criteria that pass or fail.
Binary evaluation criteria
Every criterion is pass/fail, not a scaled score. Scaled scores compound probability noise and produce unreliable results. Binary criteria are clear, auditable, and reproducible.
3 independent AI judges
Each evaluation is run by three separate AI judges independently. No judge sees the others' verdicts before scoring. This prevents anchoring bias and surfaces genuine disagreement.
Blind comparison
Judges don't know which version is the original and which is the optimized one. This eliminates recency bias and ensures the verdict is based solely on output quality.
Scoring
Your score is the percentage of binary criteria that pass across all three judges. A skill scoring 75% means 75% of criteria passed in blind evaluation.
Real compute, real stakes
Each optimization runs multiple optimization rounds with 3 independent blind judges per round. That's real API spend whether the skill improves or not. The refund guarantee exists because we're confident enough in the methodology to absorb the compute loss when it fails.
Refund trigger
You get a full refund if either condition is met:
- !Less than 10% improvement over your original skill
- !Less than 60% blind eval win rate across all judges
What makes a good eval
- →Real-world task relevance — test cases match how the skill is actually used
- →Diverse test cases — cover edge cases, not just the happy path
- →Clear success criteria — outcomes that can be objectively verified as pass or fail
Full methodology and reports
Our evaluation methodology, individual skill reports, and failure analyses are published openly on GitHub. We share the what and the proof, not the how.
github.com/willynikes2/skill-evals →How to Install a Skill
Every download bundle includes platform-specific files. Pick your platform and follow the steps below. The full bundle also includes an INSTRUCTIONS.md with the same guide.
Claude Code (CLI)
- →Save the .claude.md file to your project root as CLAUDE.md, or drop it into .claude/skills/ for project-scoped rules.
- →Alternatively, add mcp-config.json contents to your Claude Code MCP settings (~/.claude/settings.json).
Claude (Web / API)
- →Copy the contents of the .prompt.md file into your system prompt.
- →For Claude Projects: open Project Settings → Instructions and paste it there.
Cursor
- →Save the .cursorrules file to your project root as .cursorrules.
- →Or paste the contents into Cursor Settings → Rules for AI for global rules.
Windsurf
- →Save the .windsurfrules file to your project root as .windsurfrules. Windsurf picks it up automatically.
ChatGPT
- →Go to Settings → Personalization → Custom Instructions and paste the .chatgpt.md contents.
- →For a Custom GPT: open Configure → Instructions and paste it there.
Gemini
- →Save the .gemini.md file to your project root as GEMINI.md. The Gemini CLI loads it automatically.
Any other AI
- →Use the .prompt.md file — it's a universal format with no platform-specific markup.
- →Paste it into any AI assistant's system prompt, custom instructions field, or directly into the chat.
What's in every bundle
.claude.mdClaude Code format with YAML frontmatter.cursorrulesCursor editor rules format.windsurfrulesWindsurf editor rules format.chatgpt.mdChatGPT custom instructions format.gemini.mdGemini format — save as GEMINI.md.prompt.mdUniversal format for any AI assistantmcp-config.jsonMCP server configurationINSTRUCTIONS.mdThis install guide, offline*-optimization-report.mdBenchmark results and analysischecksums.sha256SHA-256 hashes for integrity verification
Using Your Optimized Skill
Once you receive your optimized files, deploying them is straightforward. Follow the instructions for your platform below.
Claude Code
ChatGPT
Cursor
Gemini
Windsurf
Pricing
Simple, transparent pricing with no surprises.
One-time payment per skill. Get back your optimized .md, MCP config, system prompt, and report card PDF. Volume deals available with purchase history.
Re-optimize skills you already own, access all Presient-made skills, priority queue, and automatic re-optimization when models update.
Bring your own API keys, run unlimited optimizations, full pipeline access, and priority support.
Team licenses, dedicated support, custom integrations, and SLA guarantees.
List your optimized skills on the marketplace and keep 70% of every sale.
See full plan details on the pricing page.
For Creators
Built a great skill? Get it optimized, then sell it. No application. No gatekeeping.
Optimize your skill
Submit your skill and receive a benchmark-verified, improved version.
List on the marketplace
Publish your optimized skill to the Presient Labs marketplace with one click.
Keep 70%
Every time someone buys your skill, you keep 70% of the sale. Automatically.
Submission guidelines
- →Format: plain Markdown (.md) — no proprietary formats or binary attachments
- →Minimum quality bar: skill must pass our eval pipeline before marketplace listing is enabled
- →Include a short description (1–3 sentences) explaining what the skill does and which platforms it targets
- →Skills must be your original work or properly licensed for redistribution
Marketplace rules
- →No plagiarism — submitting another creator's skill as your own results in permanent removal
- →Skills must pass evaluation before going live — unverified skills cannot be listed
- →Pricing guidelines: set your own price, minimum $5, maximum $500 per skill
- →We reserve the right to remove skills that violate platform policies or misrepresent capabilities
Revenue model
- →70/30 split — you keep 70% of every sale, we take 30%
- →Monthly payouts — earnings are paid out on the 1st of each month
- →Minimum payout threshold: $20 (earnings roll over to next month if under threshold)
- →Payouts via Stripe — requires a connected Stripe account
What makes a good skill
- →Clear use case — a buyer should understand in one sentence what problem this solves
- →Tested on multiple platforms — skills that work across Claude, Cursor, and Gemini have broader appeal
- →Documentation included — explain any required setup, expected inputs, and known limitations
- →Specific over generic — a skill for "reviewing TypeScript PRs" outperforms a skill for "code review"
Learn more on the creators page.
Trust & Security
Your skills are your intellectual property. We engineered the pipeline to never compromise that.
- →
Zero retention
Your original skill is purged after optimization. We don't store what we don't need.
- →
No training on your data
Your skills are never used to train models or improve our pipeline.
- →
Private by default
Only you can see your optimized results. Marketplace listing is always opt-in.