How to use Claude to review nonprofit website changes
A nonprofit web administrator uses Claude Code with task logs, separate branches, pull requests, independent subagent review, and automated scans, while retaining human approval.
Suggested prompt
Suggested prompt written by Agents In Use, not the source author. Replace bracketed inputs and review tool permissions first.
Help maintain [nonprofit website repository]. Record planned work in TODO.md and completed work in CHANGELOG.md. Work on a separate branch through a pull request. Follow CLAUDE.md. Have an independent subagent review changes using [minimal threat model] and [focus areas], then verify its findings. Keep scope expansions as separate todos. Implement appropriate security checks and axe accessibility scanning, and leave final approval to me.
Reach people exploring AI agents
See placements →Evidence at a glance
Sourced account- Source poster / publisher
- bradknowles
- Original account
- Read the source ↗
- Published
- Source link checked
- Reported result
- The author reports that Claude implemented security-enhancing GitHub Actions tools and WCAG scanning with axe. Pushes undergo scans that block builds when problems are found, and the author considers the work examined so far an improvement.
- Human involvement
- The author defines repository rules and review constraints, examines the work, and remains the human gate for approving all changes.
A linked account supports attribution. It does not mean we independently reproduced its result.
Report an issue with this case
Open correction emailOpens your email app with the case link included. Nothing is sent automatically.
How the agent did it
- Record planned nonprofit website work in TODO.md before starting changes.
- Commit work on a separate branch and handle changes through pull requests.
- Have an independent subagent examine each round using a minimal threat model and specified focus areas.
A volunteer nonprofit web administrator reports using Claude Code to improve a website while keeping human approval over changes. The workflow carries forward lessons from an earlier job, where the author used Cursor with Claude for Ansible development and recorded corrections in CLAUDE.md.
For the nonprofit work, tasks are recorded in TODO.md before execution, and completed work goes into CHANGELOG.md. Changes proceed through pull requests, with work committed on separate branches as it progresses. Local copies, the repository, and private cloud backups provide copies of the work.
The author also has a separate subagent check each round of changes. That reviewer receives a minimal threat model and specific areas to examine, without the parent's local context. Findings return to the parent for verification; confirmed problems enter the next repair round. Work that would expand an existing pull request's scope becomes a new todo instead.
Claude has implemented security-focused GitHub Actions tooling and WCAG scanning with axe. The author reports that scans run on every push and block builds when problems are found. They remain the human approval gate and describe the changes examined so far as improvements.
Who it's for: Website maintainers who want agent assistance while retaining explicit approval and documented review.
Prompt
Suggested prompt (written by Agents In Use, not the author's)
Source: Hacker News · bradknowles
Frequently asked questions
How does the independent agent review work?
A subagent without the parent's local context receives a minimal threat model and focus areas. It reports findings to the parent, which verifies them before confirmed problems enter the next repair round.
Who approves the website changes?
The author explicitly remains the human gate for approving all changes because they do not yet trust the tooling enough to relinquish approval.
What automated checks did Claude implement?
The author reports security-enhancing GitHub Actions tools and WCAG scanning with axe. Scans run on every repository push and block the build if problems are found.
More details
Remaining steps
- Verify reported findings in the parent agent and schedule confirmed problems for the next repair round.
- Run security and WCAG scans on repository pushes, with the author retaining approval over changes.
About this account
Claude implements changes and automated checks, while a separate subagent reviews each round and the parent verifies findings before further repairs.
Platform: Hacker News. Attribution: https://news.ycombinator.com/item?id=50411009. Prompt: Suggested (written by Agents In Use, not the author's). Sourced by the Agents In Use pipeline.
Credit: bradknowles on Hacker News
Source published
Added to Agents In Use
Source checked
Tools and topics
Claude Code · GitHub Actions · axe · CLAUDE.md · TODO.md · CHANGELOG.md
Website maintenance · Code review · Accessibility · Security · Nonprofit
Editorial completeness: 100/100. Measures documented fields, not agent performance or verified outcomes.