Build Ansible automation with Claude, recording lessons in CLAUDE.md
The author used Cursor and Claude to write nearly all Ansible code, auto-updating CLAUDE.md with lessons learned and logging plans to files to survive compaction; also uses Claude Code for a nonprofit website. ·
Done withCursorClaudeClaude Code
- What they did
- At a previous job, the author used Cursor with Claude for nearly all Ansible work, leaning on Puppet and Chef experience. Every lesson learned went into CLAUDE.md automatically, and Claude wrote its findings and plans to local files before each step so it could resume after compaction. For the nonprofit site, work goes through TODO.md, CHANGELOG.md, branches and PRs, with a context-free sub agent reviewing each round.
- How it went
- The author says the work looks like a solid improvement over what existed before. Push-time scans (security Actions, axe for WCAG) block bad builds. The author still approves every change personally and doesn't yet trust the tooling enough to loosen that.
- Worth knowing
- Cap the number of review rounds and require fewer fixes each round, and record out-of-scope findings as new todos, so PRs converge instead of growing.
Try it yourself with Cursor
Help me build Ansible automation in this repo to [goal, e.g. patch and configure my servers]. Before starting, create a CLAUDE.md and add a short lesson to it each time you hit a mistake or learn a convention. Write your plan to plans/[task-name].md and keep it updated as you go so nothing is lost when context is compacted. Run in check mode and show me the results before applying any change to real hosts.
Want help getting this working? Join the Vibe Coding Academy ↗ Run by Alex Finn, who also runs this site.
Source: HN · Nov 8, 2026
Related use cases
Build a C compiler with a team of parallel agents
Anthropic's engineering write-up on tasking Opus 4.6 agent teams with building a C compiler.
Build a thyroid episode warning model from wearable data
Claude Code was given 9.5 years of Apple Watch and Whoop data and built an ML model that backtested as warning of a prior episode before labs confirmed it.
Generate custom interactive HTML lessons for students
A teacher gives Claude Code a passage, level, vocab list and goals, and gets back a full lesson file with flip cards and comprehension questions, replacing hours of weekly worksheet work.
Review Linux kernel patches
Google engineers launched Sashiko, an agentic AI code reviewer for the Linux kernel.
Custom agent
Build a full CRM with email, video and calendar
A community member used Replit Agent over a couple of weeks to build a CRM with email integration, video sales letter recording and meeting scheduling.
Try Devin for pull requests on a company codebase
A commenter's team tried Devin but found PRs slow and still needing review, so it felt like a junior developer.
Get 5 like this every morning
The best things people got an AI agent to do, each with the prompt to try it.