FlyPython: Learn to ship Python with AI coding agents

English · 中文 · 🌐 Online Portal

Real projects; your agent does the typing, verify.py decides when you're done.

Challenge courses with objective verification. Pick a folder from courses/, solve the contract in TASK.md with your coding agent as the tool, and prove it with verify.py — which prints a claim code per checkpoint. Prefer a guided path? COURSE.md still runs an agent-taught mode, and paths/ sequences courses into badge routes. Continue on flypython.com.

FlyPython is a practical, bilingual repository for writing good Python and turning it into products people can rely on. It combines AI-coding methods, task playbooks, runnable examples, reusable templates, and reviewed primary sources for APIs, automation, agents, Skills, and MCP.

It is not a beginner link dump. The goal is to help you move from “the agent wrote code” to “a user outcome is verified.” The repository owns the reviewed source content and data; flypython.com turns pinned versions into a browsable learning experience.

Current release: v0.1.2 — see CHANGELOG.md; release and delivery status is tracked in the repository-to-website operating model.

Start in three minutes

You need a coding agent that can run commands and reach the network (Claude Code, the Codex app, Cursor, GitHub Copilot in VS Code, DeepSeek Harness, Kimi Code, or ZCode — chat-only web AIs cannot run these courses). Install the FlyPython Skill in your agent, then paste this one sentence:

Read https://flypython.com/skills/flypython/SKILL.md and start the FlyPython course da-eda.

The agent authorizes you with a one-time link (you never hand it a password), fetches the course files itself — you download nothing — and drives the challenges with you. New to driving an agent? Start with the tool course for your agent: Hands-on Python with Claude Code, Codex app, Cursor, GitHub Copilot, DeepSeek Harness, Kimi Code, or ZCode.

Maintainers can still reproduce an example locally without any agent:

git clone https://github.com/flypythoncom/python.git
cd python
python examples/product-slug/verify.py starter --expect-failure
python examples/product-slug/verify.py solution

Verify and record a course

In the course folder, run python verify.py after each change. It tests only starter/ and reports every checkpoint:

• [open] — the task is unfinished.
• [passed] — the tests pass and verify.py prints a claim code.
• [pending] — a reflection checkpoint: answer that lesson's questions, then run python verify.py --attest l01 (repeat for each completed reflection) to get its [attested] code.

Submit only [passed] and [attested] codes through your agent or the dashboard. check --json is the versioned v2 interface; progress --json remains for older signed-receipt integrations.

Choose what you need to accomplish

The working loop is simple: define the user outcome, inspect the real context, make the smallest testable change, verify behavior, review side effects, and record what remains unverified. AI accelerates the loop; it does not replace engineering judgment.

Prefer a guided reading path and ongoing updates? Continue on flypython.com. The repository remains the inspectable source; the website helps you choose the next useful step.

For AI Agents and LLMs

If you are an LLM agent or coding assistant (Cursor, Windsurf, Claude Code, Copilot, Perplexity):

• Ingest repository index: llms.txt or catalog.json
• Universal IDE rules template: templates/AGENT_RULES.example.md
• Machine-readable manifest: content-manifest.json
• Official web knowledge dump: https://flypython.com/llms-full.txt

Reviewed source catalog

Four maintainer-reviewed learning paths cover foundations, web and APIs, automation, and AI agents:

33 reviewed resources · Catalog reviewed 2026-09-06 · 25 intermediate or advanced · Primary sources first

• Python foundations — Learn the language, environments, dependencies, typing, and tests that reliable Python work depends on. (8 resources)
• Web and APIs — Build typed services and applications that connect Python logic to users and other systems. (7 resources)
• Automation — Turn repeatable work into maintainable scripts, browser workflows, and data pipelines. (7 resources)
• AI agents — Learn tools, structured output, state, evaluation, and the safety boundaries of agent systems. (11 resources)

Full tables — why each source is included, its level, and its access and risk notes — live in the catalog README.

Missing an important official source? Propose a resource or report a correction.

What this repository owns

• First-party bilingual guides and task playbooks.
• Runnable, verifiable Python examples and reusable agent-work templates.
• Reviewed official documentation, standards, and project sources.
• Deterministic manifests, validation, exports, and safe link audits.
• A human-review queue for the Python Project Radar.

The website owns presentation, navigation, search, newsletter, and future paid experiences. It consumes a deliberate pinned version of this repository; it must not silently fork or rewrite the source claims. Maintainers can follow the repository-to-website operating model to add measurable calls to action without inventing unavailable products.

Repository structure

assets/                README images (verify.py terminal demo)
catalog/
  README.md            browsable reviewed catalog (English, generated)
  README_cn.md         browsable reviewed catalog (Chinese, generated)
  catalog.yml          catalog status and review date
  paths.yml            bilingual learning-path definitions
  resources/           one reviewed resource per YAML file
  projects/            human-review queue for current Python projects
courses/              challenge courses with TASK.md contracts and verify.py claim codes
paths/                learning paths sequencing courses into badge routes
guides/               Python engineering and AI-coding methods
playbooks/            repeatable task procedures and definitions of done
examples/             small runnable projects with automated verification
templates/            task, plan, review, and verification starters
schema/
  *.schema.json       versioned machine-readable contracts
catalog.json         deterministic public export for consumers
radar.json            deterministic Project Radar export for consumers
content-manifest.json versioned paths, summaries, and checksums for the website
tools/                generation, validation, example, and link-audit commands
tests/                content consistency and behavior tests
docs/                 consumer contract and curation policy

catalog.json and content-manifest.json are generated; do not edit them by hand. Website consumers read both from a pinned commit, verify checksums, and record that revision in their own lock file. They must not fetch a moving branch during a production build.

Example immutable URL:

https://raw.githubusercontent.com/flypythoncom/python/<full-commit-sha>/catalog.json

The content manifest lets flypython.com render the matching bilingual guide or playbook without owning a second editable copy.

Contribute

Propose a resource, report a correction, or improve a course. Read CONTRIBUTING.md first — it covers the local setup, the deterministic checks CI runs, and how to regenerate the generated exports. Also read the curation policy before proposing a resource or changing its classification, and the consumer contract for website integrations.