
Un-Default
io.github.NomadBuilderPublished by GitHub account NomadBuilderv0.5.6Updated Oct 24, 2026
Catch racist, sexist & ableist defaults in AI and human copy. Local rules, no account.
Overview
Reviews copy, docs, UI strings, and live pages for racist, sexist, and ableist defaults, explaining findings and suggesting clearer alternatives.
- What you can do
- Un-Default scans text for biased defaults, sexist assumptions, ableist metaphors, and documented dogwhistles, then explains what it noticed and offers clearer alternatives. Its MCP tools are un_default_scan, un_default_fix, and un_default_feedback. Matching runs on local open rules and does not require an AI model. It can also scan paths, changed files, and batches of URLs, and output JSON or SARIF.
- When to use it
- Use it after an assistant writes UI copy, documentation, or marketing text, or before publishing, to catch problematic defaults. It also fits into pull request checks and editor workflows where copy should be reviewed consistently.
- Requirements
- Runs locally as an npm package via npx (Node.js required). No account, API key, or environment variable is declared. The MCP server is started with the mcp subcommand. Network access is only needed for URL scanning or optional anonymous action pings.
Installation
In SourceWeft
- Open Un-Default in the dashboard and add it to a workspace.
- Enable the server for the chats that should use its tools.
Desktop only via STDIO. STDIO servers start a local process, so they need the SourceWeft desktop host.
Other MCP clients
Follow the launch instructions in the repository.
README
Un-Default
One thoughtful-language check for every way you build.
Formerly Anti-Default. Same tool — the name shifted so it reads as undoing defaults, not opposing people. Old URLs under
darkai.ca/anti-defaultredirect here.npx anti-defaultstill works as an alias fornpx -y anti-default.
Un-Default reviews copy, docs, UI strings, and live pages for racist defaults, sexist assumptions, ableist metaphors, and documented dogwhistles. It explains what it noticed and offers clearer alternatives.
No account. No AI required for matching. Open rules you can tune.
After Claude writes UI copy · Try the web app · Add the Chrome extension · Star on GitHub
After Claude / Cursor writes UI copy
One command — skill + MCP + PR check. This is the habit that sticks:
That also writes examples/un-default-demo.md and prints a sample scan so you see a finding immediately. Browser demo (shareable): /?demo=1.
Or install the Claude Code plugin from this repo’s marketplace:
Paste-ready MCP JSON and Claude Project instructions:
https://darkai.ca/un-default/for-agents/
Directory checklist: docs/PUBLISH-DIRECTORIES.md
Paste MCP config (Cursor · Claude Desktop · Claude Code)
Tools: un_default_scan · un_default_fix · un_default_feedback
Add it in under a minute
That is the whole local setup. It prints findings and exits non-zero on clear hits. Ambiguous or quoted matches stay advisory.
To pin it in a team project:
Want the complete project setup?
init adds an ignore file, changed-files GitHub workflow, Cursor + Claude
skills, .cursor/mcp.json + .mcp.json, Claude/Cursor after-edit hooks,
CLAUDE.md instructions, and inclusive-check / inclusive-fix scripts.
Existing files are never overwritten. Hooks surface hard findings back to the
agent after Edit/Write so Claude keeps checking even when nobody remembers.
AI-copy habit (definition of done)
Agents should not mark UI/docs work done until hard findings are cleared or explicitly marked fine in context. Soft/coded hits stay advisory.
MCP server
See for-agents for host-specific paste steps.
Scan the way you work
Ignore what doesn’t matter
Commit a .undefaultignore so day-two noise doesn’t drown the team (example):
GitHub Action
Keep racist, sexist & ableist defaults out of the PR loop. This scans only changed files, fails on clear new findings, and leaves a checklist comment:
Prefer SARIF for Code Scanning? Set format: sarif and output-file: un-default.sarif. Full inputs: action.yml.
Anonymous Action pings (no repo name or findings) feed public totals at darkai.ca/un-default/usage. Badge:
Agent skill
npx -y anti-default init installs .cursor/skills/un-default/SKILL.md — the
definition-of-done workflow for AI-generated copy: fix → scan → ask or
feedback → re-scan.
Source: skills/un-default/SKILL.md ·
feedback model: feedback/README.md
Use it as a library
The npm package has no runtime dependencies and exposes the same analyzer used by the CLI and web app:
Chrome extension
Same rules as the app — on the page you’re already looking at.
Runs offline from a bundled rule list. No tracking. No AI calls.
Dev / unpacked: npm run extension:pack → Load unpacked → extension/ · details in extension/README.md
Web app — try, learn, tune
- Review — paste, upload, or scan a URL; export Markdown, CSV, or a checklist
- Swap — turn one phrase into clearer alternatives
- Dogwhistles — learn what a phrase can signal and when context matters
- Rules — tune the open rule catalog for your project
Why teams reach for it
- Before publish — catch defaults in marketing, docs, and product copy
- In the PR — Action + SARIF/JSON so machines and humans both get a signal
- On the live page — extension for writers who never open a terminal
- Without a lecture — suggestions are invitations; context always wins
Every suggestion cites the style guides and references behind it → /sources
Customize & self-host
- Tune in the UI: /rules · edit code:
src/lib/rules.ts - Practice corpus:
npm run corpus - Static export:
npm run build→out/
Subpath (DarkAI):STATIC_EXPORT=true BASE_PATH=/un-default npm run build
Production lives at darkai.ca/un-default and is also vendored in DarkAI.
License
Source: README.md at commit 8d707ee
Tools
0Version history
1- v0.5.6LatestSep 16, 2026
Score
50% of the applicable checks met
How complete this listing is, computed from the registry entry, the endpoint and public sources. Not a rating of the server and not a statement that it is safe or working today: reachability is a dated check, and no installs or reviews go into the score.
- Reached in the last 30 days25 pts
Keep the endpoint answering the MCP handshake; SourceWeft probes it every 30 days.
- Auth declared10 pts
Declare the credential headers in the registry entry, or keep the endpoint answering the probe so its auth can be classified.
- Tools listed15 pts
Answer tools/list without a credential, or ship a package that starts in a sandbox without one.
- DescriptionRequired15 pts
- Logo10 pts
- Documentation15 pts
- License10 pts