Prompt Enhancement Skill turns a stated outcome into a compact, paste-ready prompt that preserves the context, constraints, permissions, and success checks that affect the result.
It creates, repairs, critiques, and adapts prompts for:
- coding and action-taking agents
- research and structured extraction
- interface, image, video, and audio generation
- browser automation and computer use
Pasted prompts, files, and examples are treated as data—not instructions. Secrets and sensitive values are replaced with named placeholders.
npx skills add pymodel/prompt-enhancement-skill@prompt-enhancement-skillInstall with the skills CLI, then invoke it by asking for a prompt explicitly.
Write a prompt for a coding agent to add CSV export to the reports page.
Create a design brief prompt for a responsive analytics dashboard.
This edition incorporates current official guidance for Claude Fable 5 and Claude Opus 5.
| Target | Prompt controls included |
|---|---|
| All current Claude models | Clear outcomes, context and motivation, structured examples, XML boundaries, explicit tool use, and long-context grounding. |
| Claude Fable 5 | Clear assessment-versus-mutation boundaries, progress claims grounded in actual tool output, scoped long-running work, and safe parallel delegation. |
| Claude Opus 5 | Complete task specifications up front, explicit response-length control, narrow scope, controlled subagent use, and no redundant self-check loops. |
The skill does not encode volatile API defaults as prose. Model effort, thinking mode, and harness settings remain platform configuration and should be checked against the linked official docs.
Generic “make it premium” prompts produce generic interfaces. Prompt Enhancement Skill instead requires a visible design contract:
- product outcome, audience, primary task, required content, and non-happy-path states
- hierarchy, layout, typography, palette, surfaces, imagery, density, and responsive behavior
- a context-specific visual direction—not copied branding or default component patterns
- explicit animations and interaction states, with reduced-motion and accessibility behavior
- named existing tokens, components, and assets to preserve
- browser or screenshot checks at specified breakpoints when an agent implements the design
For Claude, the interface template adds visual-direction controls for distinctiveness, high-impact interaction moments, and visual verification.
- Resolve the target, objective, trusted input, output contract, constraints, and success conditions.
- Route to the smallest relevant template and product profile.
- Compose only instructions that change the likely result.
- Bound agents with editable scope, verification, and confirmation gates for irreversible actions.
- Deliver one copyable prompt, with assumptions or setup only when material.
.
├── SKILL.md # Core operating contract and routing
├── references/
│ ├── patterns.md # Prompt repair diagnostic table
│ ├── templates.md # Task-specific prompt templates
│ └── tool-profiles.md # Current product and Claude 5 guidance
└── LICENSE