Mark HollandSenior AI Solutions Engineer

Prompt, Skill & Agent Library

ProductionBuilt at work for a national healthcare data company

Good prompts tend to disappear into chat histories. This library gives every team one place to keep them, along with Skills, skill chains, and agents. Colleagues vote, the most helpful rise, and each one downloads straight into the AI apps and IDEs people already use.

The Prompt Library catalog: a search box, category chips by team, and the highest-ranked prompts first, such as Write test cases from acceptance criteria.

What it does

  • Company-wide library of prompts, Skills, skill chains, and agents by team and role, with peer voting to keep the most helpful, downloads for major AI apps and IDEs, and AI adoption training.

From my resume.

Who it is for

Every team and role across the company, from people new to AI to the ones writing agents.

How it works

Prompt, Skill & Agent Library, step by step
  1. Submit

    Anyone submits a prompt, Skill, skill chain, or agent.

  2. Screen

    Submissions are screened automatically, and anything that looks like PHI or PII is held for review.

  3. Vote

    Peers vote, and the most helpful rise.

  4. Download

    Downloads for major AI apps and IDEs.

  5. Teach

    AI adoption training builds on the same library.

The real screens

The Prompt Library catalog: a search box, category chips by team, and the highest-ranked prompts first, such as Write test cases from acceptance criteria.
The catalog, organized by team, with the highest ranked first. Shown with demo data.
Ranked prompt cards in two team sections, each with tags, a star rating, the number of ratings, and how often it was used.
Peer ratings and use counts decide what rises. Shown with demo data.
Prompt Library submission held for compliance review: the gate lists a name with a date of birth, a date of birth, an SSN-like value and a member ID, and the draft is kept below with its title and body.
  1. Held, not lost. The gate lists what it found and keeps the draft for the author to fix.

Synthetic test dataA submission with patient identifiers is held for review, and the draft is kept. The patient here is a placeholder. Shown with demo data.
Prompt Library moderation: example coverage, 6 of 30 prompts ready, and one compliance hold waiting for a decision to clear and publish, force publish as a recorded override, or keep it held.
A person clears, overrides, or keeps each hold, and an override is recorded. Shown with demo data.
A Skill page, Acceptance criteria to test cases, with its skill.md and a Provenance section linking back to the prompt it was promoted from.
Every Skill keeps a link to the prompt it grew from. Shown with demo data.
The optimization page: queues of prompts and skills to improve, and recent optimizer runs waiting for a person to accept or reject them.
Optimizers propose; a person accepts before anything becomes current. The scores shown come from the app's mock optimizer, not a measurement. Shown with demo data.

The decision that matters

Risky drafts wait for review

When the gate finds something that looks like PHI or PII, the submission is held for review and the draft is kept, so the author can fix it and resubmit.

Checked evidence

  • Verified Submissions are screened automatically and held for review if regulated data is detected; the draft is preserved.