Prompt Architect

A Claude Skill that helps you build new prompts or improve existing ones — for a non-technical audience, using a plain-language framework covering context, inputs and tools, guardrails, output format, and self-verification.

It leads with a gap analysis on existing prompts (not an immediate rewrite), asks intake questions before drafting new ones, checks Anthropic's official docs before recommending a model, and audits its own drafts for token bloat before handing them over.

Installation

Claude.ai, Claude Desktop, and Cowork

1. Download this repository as a zip: Code → Download ZIP on this page (or git clone it).
2. In Claude, go to Settings → Capabilities → Skills → Upload skill.
3. Select the downloaded zip. SKILL.md is at the root of this repo, so no repackaging is needed.
4. Enable the skill.

This upload is per-account. If you're on a Team or Enterprise plan, you can share it with your organization from the same settings screen instead of everyone uploading individually.

Claude Code

git clone https://github.com/dhand-personal/prompt-architect.git ~/.claude/skills/prompt-architect

Or, for just this repo/project:

git clone https://github.com/dhand-personal/prompt-architect.git .claude/skills/prompt-architect

Restart your session after installing.

Usage

Ask Claude to review or build a prompt, directly:

Can you look at this prompt and tell me if it's any good?
[paste your prompt]

I want a prompt for an AI that [does X]. Can you help me build it?

What Claude model should I use for [task]?

The skill triggers automatically on requests like these — no special syntax needed.

How it works

Every prompt is checked against six components: Role & Context, Inputs & Tools, Task & Success Criteria, Guardrails, Output Format, and Self-Verification & Efficiency. See SKILL.md for the full framework and behavior rules.

Test cases covering the main behaviors — gap analysis, new-build intake, model-tier sequencing, and the token-efficiency check — are in evals/evals.json.

Version history

• 1.0.0 — Initial release. 6-component framework, gap-analysis-first review path, build-from-scratch intake path, review checkpoint before treating any draft as final, live official-docs check for model-tier recommendations, and a token-efficiency pass on both drafted prompts and the skill's own self-verification checklists.

License

MIT