sleekdotdesign/agent-skills: The Instruction Manual for Your AI Designer

Moving beyond raw prompts to standardized, agent-native skills that turn LLMs into professional mobile architects.

• View on GitHub • More from sleekdotdesign

A vintage glass vacuum tube being slotted into a mechanical head, containing lines of perfectly formatted Markdown text instead of wires. — The new SDK is written in Markdown.

Key Takeaways

The repository replaces traditional client libraries with Markdown-based instructions that agents read as procedural intelligence.
A standardized SKILL.md file uses YAML and strict rules to teach AI agents how to manage REST API calls and asynchronous states.
The skill enforces a visual feedback loop by requiring agents to poll for job completion and capture screenshots of their design work.
This modular approach allows developers to install domain-specific capabilities into agents via simple CLI commands rather than manual prompting.

The End of the Client Library

Software development is undergoing a quiet but radical shift. For decades, integrating a third-party service meant installing a heavy client library and writing glue code. Today, AI agents are replacing human glue code, but they still need to know how to talk to external systems.

Enter sleekdotdesign/agent-skills. Instead of an NPM package filled with JavaScript functions, this repository provides procedural intelligence. It is a library of standardized instructions formatted in Markdown.

How the agent only loads the skill into context when a relevant design intent is detected.

By installing this skill, a general-purpose AI like Claude Code or GitHub Copilot instantly learns how to orchestrate Sleek's professional mobile design architecture. The agent reads the instructions, understands the API boundaries, and executes the workflows autonomously.

Inside the Agentic Brain

The core of the repository lives in a single file: SKILL.md. This is not documentation for a human. It is the literal brain module for the AI agent.

The file begins with YAML frontmatter that defines the skill's name and triggers. Below that, it outlines strict rules of engagement. The AI is taught how to construct valid REST API calls, handle idempotency keys to prevent duplicate billing, and manage asynchronous state.

| Skill | Description | | ----- | ----------- | | design-mobile-apps | Design mobile apps, create screens, and manage Sleek projects with AI |

Project README, Repository documentation · sleekdotdesign/agent-skills README

A mechanical eye looking at a stopwatch, glancing at a blueprint that is slowly drawing itself. — The skill teaches the agent how to patiently poll for asynchronous design completion.

Seeing is Believing

Designing user interfaces blindly is a recipe for hallucinations. The Sleek skill solves this by enforcing a strict visual feedback loop.

The skill explicitly instructs the agent to capture screenshots of newly created or updated screens. The agent submits a design request, polls the asynchronous job status until completion, and then fetches the visual result. This loop allows the AI to literally see its work and present it immediately to the user for iteration.

The asynchronous request, poll, and screenshot cycle defined within the skill.

Prompting vs. Skill-ing

The traditional approach to AI integration relies on copying and pasting massive system prompts. This is brittle, token-heavy, and inconsistent across different agent environments.

Raw System Prompts	Agent Skills (SKILL.md)
Brittle and prone to context degradation	Standardized declarative format
Requires manual copy-pasting into chat	Installable via `npx skills add`
No native tool-use constraints	Explicit API boundaries and scopes
Hard to version control and share	Versioned alongside the repository

The sleekdotdesign/agent-skills repository represents the maturation of AI workflows. We are moving from shouting instructions at a generalist chatbot to installing certified, domain-specific modules. The SDK is dead. Long live the Skill.