Challenge courses
You solve the challenge. verify.py decides.
Start with one sentence — your agent fetches the course files, you work the task contract together, and python verify.py decides every checkpoint. No downloads, no video player, no opinions about “done”. Prefer guidance? COURSE.md runs an optional agent-taught mode. Every course ships English and Chinese together.
Needs a coding agent that can run commands and reach the network — every course page carries its own one-sentence entry; you never download anything.
Start here
Your agent's tool course
Six free courses, one per coding agent. Lesson 1 is always the same three steps: get the tool running → install the FlyPython Skill (network on) → let the agent fetch the course.
Hands-on Python with Claude CodeNew
A challenge course that takes you from a broken script to a tested, verified Python report tool — one sentence starts it in your agent, which fetches the course itself. Task contract, bounded change, and objective pass/fail evidence; guided agent-taught mode included.
02Hands-on with the OpenAI Codex AppNew
The same verified Python workflow as the Claude Code course — task contract, bounded change, objective verify.py evidence — taught hands-on in the Codex desktop app: threads, approvals, AGENTS.md, and diff review.
03Hands-on with CursorNew
The same verified Python workflow as the flagship course — task contract, bounded change, objective verify.py evidence — taught hands-on in Cursor: Agent/Ask/Plan modes, .cursor/rules project rules, and accept/reject diff review.
04Hands-on with DeepSeek HarnessNew
The same verified Python workflow as the flagship course — task contract, bounded change, objective verify.py evidence — taught hands-on in DeepSeek Harness: plugin-composed agents and an auditable append-only trajectory.
05Hands-on with Kimi CodeNew
The same verified Python workflow as the flagship course — task contract, bounded change, objective verify.py evidence — taught hands-on in Kimi Code: built-in explore/plan/coder subagents keep reading, planning, and editing in separate lanes.
06Hands-on with ZCodeNew
The same verified Python workflow as the flagship course — task contract, bounded change, objective verify.py evidence — taught hands-on in the ZCode desktop workspace: Goal Mode task decomposition and the Git state as your review surface.
Skill-driven
Every course is one sentence away
Paste the entry sentence from a course page; your agent handles files, verification, and submission. Read the challenges online first — no account needed.
Difficulty: Beginner — first agent or first Python project; Intermediate — comfortable running an agent and reading code; Advanced — multi-step builds with real verification.
One source of truth for agent rulesNew
Stop maintaining diverging AGENTS.md, CLAUDE.md, and .cursorrules files — build a checker that proves your repository has exactly one rule source.
02From “it runs” to “it ships”: verifying AI-generated codeNew
Build the release-evidence machine your AI-written project is missing — a ship check that parses what actually ran, refuses zero-test projects, and writes an honest delivery record.
03Give your agent tools with MCP (Python)New
Build a stateless Model Context Protocol tool server in pure Python — JSON-RPC 2.0 dispatch, schema validation, error isolation, and the 2026-07-28 input_required round-trip.
04Exploratory Data Analysis with an AgentNew
A challenge course — point your agent at a messy shop export and produce a verified EDA summary whose numbers are checked against ground truth.
05Data Visualization with an AgentNew
A challenge course — turn a clean sales dataset into three spec'd charts plus a summary.json whose numbers are verified against the data.
06From Analysis to Report with an AgentNew
A challenge course — turn a verified EDA summary into report.json + report.md where every number traces back to its input unchanged.
07pytest in the Loop — Tests That Catch Bugs ProNew
A challenge course — given a correct library and a seeded buggy variant, write a pytest suite (fixtures, parametrize, raises) that passes on the reference and fails on the bugs. A suite that cannot catch a bug is not a suite.
08Package and Ship a CLI ProNew
A challenge course — turn a Markdown-indexer script into an installable package: pyproject metadata, console entry point, built wheel and sdist verified by the contract suite — including a real install into a fresh virtualenv.
09A Verified Web API ProNew
A challenge course — a FastAPI snippets service violates its own spec in six places. Fix the app until the provided contract suite is green, then write the tests for what the spec leaves unpinned and prove they catch planted regressions.
How it works
One sentence. One command. One honest verdict.
1. Authorize. Your agent shows you a one-time link and a short code; you compare and approve. No passwords, no downloads.
2. Solve. The agent fetches the course folder and works the TASK.md contract with you. Want a guide? COURSE.md runs an optional agent-taught mode.
3. Verify. python verify.py runs the suites, prints checkpoint status and your claim codes, and exits non-zero until the gates pass — the same honest verdict for everyone.