google-labs-code/stitch-skills

Source
GitHub
First trending
Category
Skills
GitHub stars
9,316
Main language
TypeScript
Website
stitch.withgoogle.com/docs/mcp/setup (opens in a new tab)

This page introduces an external open-source repository. It is not an HDATF product.

google-labs-code/stitch-skills

What it does

This repository packages agent instructions and helper scripts for Google Stitch. It follows the Agent Skills format and separates design, build, and utility plugins. The workflows cover generating and editing screens from text or images. Another workflow extracts standalone HTML and design rules from existing frontend code before uploading them to Stitch. Build skills describe converting Stitch screens into React or React Native components. The React workflow separates component code, hooks, mock data, and visual tokens. Its supplied syntax-tree validator checks for a Props interface and hardcoded hexadecimal colors in class attributes, not every architectural rule in the instructions. These workflows depend on configured Stitch MCP access and credentials, and the README states that this is not an officially supported Google product.

License

Apache-2.0 Permissive, with a patent grant. Commercial use is allowed; keep the notices and state your changes.

More in this category

  • NVIDIA/SkillEvaluator
    SkillEvaluator checks agent skills, meaning folders of task instructions and supporting files, and evaluates their contribution to task outcomes. Tier 1 checks structure, quality, security and related static properties. Tier 2 uses embeddings to find repeated guidance within a skill and overlap across a supplied skill catalog. Tier 3 runs an agent on the same cases with and without the skill through the Harbor evaluation framework. Reports compare the two conditions and retain coverage and incomplete evidence rather than treating missing checks as passes. The inspected comparison implementation uses shared scored cases for each metric and excludes cases judged by different models for judge-based metrics. Static quality alone does not establish safe execution or a benefit in live tasks, and a generated starter case has limited coverage. Live evaluation needs model access and an execution environment and may incur charges; the project describes its support level as experimental.
  • arjun988/blender-skills
    Blender Skills is a collection of task instructions and references for coding agents that work with Blender, rather than a standalone 3D generation model. It addresses the need to carry an artist's request through modeling, materials, lighting, optimization, and export without treating each step as an unrelated question. Inputs include a text brief, a reference image, the target engine, and polygon and texture budgets. The blender-director instruction selects specialist skills and defines a production brief before scene operations begin. Execution relies on the separate BlenderMCP addon and MCP server, with MCP meaning the interface through which an agent calls external tools. The intended outputs are Blender scene changes, rendered checks, exported assets, and a validation report, while reference matching uses camera alignment and repeated screenshot comparison. The setup guide lists Blender 3.0 or later, Python 3.10 or later, uv, a running addon connection, and a compatible agent client as prerequisites, and the checked configuration connects to localhost port 9876. Quality depends on the agent and connected tools rather than the instruction files alone, and arbitrary Python execution in Blender requires backups and restricted permissions.
  • tradermonty/claude-trading-skills
    Claude Trading Skills organizes investment research and trading review for individual investors who have limited time. It addresses fragmented market checks, risk review, trade planning and journaling rather than delegating investment decisions to AI. Users supply public market CSV files, chart images, local records or data from integrations such as FMP and Alpaca. Skill instructions and Python helpers turn these inputs into market reports, position sizing calculations and structured investment records. YAML workflow manifests specify the skill order, required artifacts and decision gates. The trader-memory-core skill keeps thesis histories in local YAML files and produces postmortem reports. A starting workflow reviews market breadth and uptrend participation before a human reviews the resulting exposure posture, with no paid data API required for the basic path. It runs through Claude Skills or Claude Code, has workflow-specific Python and integration requirements, and is neither an automated order platform nor a guarantee of profitability.
  • dotnet/skills
    .NET Agent Skills is the .NET team's collection of coding-agent guidance and host-specific agents. It packages knowledge for C# development, data access, diagnostics, builds and testing as plugins rather than providing a standalone application or model. An agent reads relevant skills in the context of code, errors and development requests to choose a workflow. Actual changes are performed by the host agent and development tools, not automatically by the repository itself. Official onboarding distinguishes marketplace or individual-skill registration and reloading in compatible hosts. The inspected SDK setup skill describes project-local installation, preserving existing settings and its required host version. Skills and MCP compatibility vary by component; Copilot agent files do not become another host's native agents. Language-server integration also requires host support. A suitable .NET SDK and task-specific tools remain necessary, and installing the collection does not activate every plugin capability.
  • janestreet/agent_skills
    Jane Street's agent_skills is a collection of text instructions for existing AI agents and chatbots, not a standalone tutoring application or model. The README currently lists a Socratic tutor skill intended for users who explicitly request teaching, quizzes or practice toward mastery. Given a topic and the learner's answers, it maintains a subskill map with levels from zero to five and uses that map to choose further questions. Its instructions begin with a short diagnostic, then ask one question at a time and narrow the question when an answer reveals a misunderstanding. Progress requires demonstrated performance on more than one independent check, including recall rather than simply repeating a supplied answer. Earlier material is revisited through spaced review, and completion includes a mixed assessment with both individual subskill questions and an integrative problem. One-off explanation requests are explicitly outside the tutoring loop, and the learner can override questioning to request a direct answer. A compatible agent can read the SKILL.md directory, or a chatbot can receive its text as a prompt, so the existing model and its instruction following remain prerequisites. Session transfer is described as a compact state block that a person can save and paste elsewhere, not an implemented storage backend or independently validated measure of learning.

Only repositories in the ranked Trendshift lists are included, and the lists are used only to find candidates. We do not copy their ranks. Descriptions, licenses and star counts come from each GitHub repository. The notes are our own reading. We have not tested these projects, and a place on a trending list does not prove quality.

View on GitHub (opens in a new tab)