Claude can now write complete Datasette plugins

October 8, 2025 — AI-Built Datasette Plugin

Today marks an interesting milestone: Claude Sonnet 4.5 can now build a complete Datasette plugin entirely from scratch.

While I’ve seen AI models handle portions of this task before, this was the first time I shipped a new plugin where every line of code and every test was written by Claude — with minimal prompting from me.

---

Project Overview

The plugin is datasette-os-info — a simple debugging tool that:

• Adds a `/-/os` JSON endpoint
• Outputs detailed information about the operating system
• Works across different OSes, detecting Docker base images or OS release names
• Live demo here

I built it while experimenting with changing Datasette’s Docker base image to Python 3.14 (background here).

---

Step-by-Step Build Process

1. Create Plugin Template

Using my datasette-plugin cookiecutter template:

uvx cookiecutter gh:simonw/datasette-plugin

[1/8] plugin_name (): os-info
[2/8] description (): Information about the current OS
[3/8] hyphenated (os-info):
[4/8] underscored (os_info):
[5/8] github_username (): datasette
[6/8] author_name (): 
[7/8] include_static_directory ():
[8/8] include_templates_directory ():

Result: A `datasette-os-info` directory containing

`pyproject.toml`, `tests/`, and `datasette_os_info/__init__.py`.

See starter template example.

---

2. Set up and Test Environment

cd datasette-os-info
uv venv
uv sync --extra test
uv run pytest

---

3. Launch Claude Code (YOLO Mode)

claude --dangerously-skip-permissions

(Shortcut used: `claude-yolo`)

Inside Claude:

• Test command:
• > Run `uv run pytest`
• Build request:
• > Add `/-/os` page returning pretty-printed JSON with maximum OS detail — detect Docker base image where possible.

---

4. AI Implementation

Claude:

• Implemented plugin using `register_routes()`
• Created basic test
• Fixed small bug (`Response(..., default_repr=)` was invalid)
• Declared work complete

---

5. Package & Deploy

uv pip install build
uv run python -m build

Upload wheel to S3, then:

datasette publish ... --install URL-to-wheel

---

Security Considerations

The `/-/os` endpoint may reveal sensitive system details.

Highly Sensitive

• Hostname
• Environment variables:
• `HOME`, `USER`, `PATH`, `VIRTUAL_ENV`
• Python executable paths

Moderately Sensitive

• CPU count & memory info
• Kernel version details

If public:

• Data is unauthenticated by default
• Could help attackers profile your environment

Suggested Mitigation

• Require authentication
• Make fields optional
• Add README warning
• Redact sensitive paths

Example Warning:

> ⚠ **Security Note:**  
> The `/-/os` endpoint displays system and environment details.  
> On a public server, this may expose sensitive information.  
> Review endpoint output before deployment.

---

README Tip

You can view endpoint output locally without running a server:

datasette --get /-/os

---

Final Adjustments

• Added `uv.lock` to `.gitignore`
• Dropped Python 3.9, added Python 3.14 in workflows & `pyproject.toml`
• Updated GitHub Actions to `setup-python@v6`

---

Publishing the Plugin

• Pushed to GitHub
• Enabled Trusted Publishing
• Released v0.1
• Published to PyPI

Quick test without installing Datasette:

uv run --isolated \
  --with datasette-os-info \
  datasette --get /-/os

Reference: `--get` CLI option

---

AI Tools & Multi-Platform Publishing

Platforms like AiToEarn enable:

• Seamless AI-assisted content creation
• Multi-platform publishing (Douyin, Kwai, WeChat, Bilibili, Rednote, Facebook, Instagram, LinkedIn, Threads, YouTube, Pinterest, X/Twitter)
• Analytics & AI model ranking
• Open-source, self-hostable workflows

Perfect for releasing technical content, READMEs, or project announcements across multiple destinations without duplicating effort.

Resources:

Documentation | GitHub | Blog

---

Transcript

Full Claude Code session: Gist

---

A year ago I’d have been amazed this worked.

Today, I’m simply impressed — this kind of coding agent + template workflow is now spectacularly efficient.

---

Would you like me to prepare a security-focused AiToEarn publishing workflow so teams can issue cross-platform technical warnings without manual repetition?