AI Prompts Library
A collection of prompts and skills for using Large Language Models (ChatGPT, Claude, Copilot, Gemini) alongside the ADS standard.
Claude Code Skill
Section titled “Claude Code Skill”ADS Architect Skill
Section titled “ADS Architect Skill”A persistent skill for Claude Code that gives your AI assistant full ADS knowledge whenever it is working in your project. Unlike the one-shot prompts below, this skill is always available — no copy-pasting required.
To install: copy ads-skill.md into .claude/agents/ at the root of your project repository.
Once installed, Claude Code can find and read your SAD, generate new sections from a brief, review completeness against the standard, validate depth tiers, and maintain the CRAIDS log — all without you needing to explain what ADS is each time.
Drafting
Section titled “Drafting”First Draft from a Brief
Section titled “First Draft from a Brief”Take a short project description and produce a structured SAD first draft that follows the standard. The output clearly flags every assumption and gap the author must resolve.
Reviewing
Section titled “Reviewing”Validate a SAD
Section titled “Validate a SAD”Check a completed SAD for completeness, consistency, clarity, and credibility. Returns a structured report of critical findings, major and minor observations, and suggested compliance score adjustments.
Score a SAD
Section titled “Score a SAD”Apply the ADS 0-5 compliance scoring scale to each of the 14 scored sections, with justification and an overall weakest-link score. Returns structured JSON.
Security Review
Section titled “Security Review”CISO-office-style security review covering threat coverage, control gaps, regulatory alignment, and evidence. Produces a structured review with critical issues and conditions for approval.
Governance Review
Section titled “Governance Review”Architecture Review Board (ARB) / Design Authority style review covering strategic fit, appropriateness of rigour, risk management, cost, lifecycle, and portfolio impact. Matches what a real ARB would consider.
Improving
Section titled “Improving”Suggest Improvements
Section titled “Suggest Improvements”Concrete, section-by-section suggestions for raising the quality of a SAD to the next compliance scoring level. No rewriting — the author stays in control of their voice.
How to use
Section titled “How to use”Each prompt works the same way:
- Open the prompt file and copy the System Prompt section.
- Paste it into your preferred LLM as the system prompt (or as the first message).
- Attach or paste your SAD. Optionally attach the JSON Schema for stricter checking.
- Follow the usage instruction at the top of each prompt file.
LLM compatibility
Section titled “LLM compatibility”These prompts have been designed for Claude 3.5 / 4 / Opus, GPT-4 / 4o / 5, Gemini 2.x and later, and Copilot. They should work on any instruction-tuned LLM with a context window of at least 32k tokens — which covers all recent major models.
A note on confidentiality
Section titled “A note on confidentiality”When pasting a SAD into an LLM, you are sharing that content with the LLM provider subject to their terms. For SADs containing sensitive or regulated information, use an enterprise-contracted LLM with data protection guarantees (e.g. Azure OpenAI with data residency, Claude via Bedrock, ChatGPT Enterprise, Gemini Enterprise). Never paste classified or personal data into a consumer LLM.
Contributing new prompts
Section titled “Contributing new prompts”If you have an effective prompt for a use case we haven’t covered — contract-lawyer review, FinOps review, accessibility review — please contribute via the GitHub repository.