Learn by example
Run a support agent on the page, with and without skills, then build it yourself. No account for its first three steps.
Use a skill now
Find one, add it to Claude Code, Cursor, LangChain or your own code, and see it work. No account.
Just browse
Every skill, with its safety scan and any test results. No signup.
How it works
A skill is a file of instructions. ASKILL.md holds what an agent should know for one kind of task, such as how to handle a refund or how to use a particular tool. Your agent sees a short list of its skills and opens the one a task calls for, so its prompt stays small.
A tested skill shows its results. A benchmark runs the same test cases with the skill and without it. playwright-cli, for example, passed 23 of 23 cases with the skill and 7 of 23 without it. Every public skill also gets a safety scan, and SkillScore sums up the evidence. Results depend on the model and the cases, so treat a score as a shortlist, and test a skill on your own cases.
Your code runs your agent. DecimalAI never hosts it. Claude Code and editors read skill files from your project. In Python, your framework’s agent_config() gives your agent its skills. With no DecimalAI key, they come from files in your project and nothing is sent to DecimalAI.
What an account adds
Works with what you use
Claude Code, Cursor and other editors, LangChain, OpenAI Agents, Claude Agent SDK, the OpenAI SDK, the Anthropic API, the Gemini API, Pydantic AI, Google ADK, CrewAI, and more. All frameworks shows which way of adding a skill works on each one. These docs describe SDK 0.28 and later. On an older release, see Migrations, and the changelog for what changed when.Also in these docs
Publish a skill
Write a skill, benchmark it and share it with your team or the public registry.
Catch regressions in CI
See what a change to your agent would break, on every pull request.
SDK and CLI
The Python SDK and command-line reference.
API
Authentication, errors and the REST endpoints.