dx-roaster

Category: Developer Tools · Version: 1.0.0 · Author: OpenDirectory · Compatibility: All targets (claude, codex, opencode, gemini, anti-gravity, openclaw, hermes)

Brutally honest developer-experience audit for a GitHub repo. Scores 10 DX dimensions, writes a roast in your chosen tone, ships a prioritized fix list, and sketches what your README could look like. Use it before Show HN, Product Hunt, npm publish, or any launch where the first five minutes decide whether visitors stay or bounce. This is not a security scan or a linter. It judges the same thing a stranger would in 30 seconds: can I understand this, install it, and run it?

Install

npx CLI (Recommended)
skills.sh
Claude Code Native
Claude Desktop
Manus AI

npx "@opendirectory.dev/skills" install dx-roaster --target claude

npx skills add Varnan-Tech/opendirectory --skill dx-roaster

/plugin marketplace add Varnan-Tech/opendirectory
/plugin install opendirectory-gtm-skills@opendirectory-marketplace

Copy https://github.com/Varnan-Tech/opendirectory/tree/main/skills/dx-roaster into download-directory.github.io to download the ZIP.
Open Claude Desktop -> Customize -> Skills -> + -> Upload a skill and drop the folder containing SKILL.md.

How It Works

You point the skill at a repo (local path or GitHub URL) and pick a tone.
The agent scans the README, install steps, code samples, docs, examples, badges, community files, license, and recent activity.
It scores 10 DX dimensions on a 0–10 scale.
It writes four files: a shareable roast, a prioritized fix list, an ideal-README sketch, and raw scores.

The whole thing runs in under a minute. No external services. Read-only on your repo.

The 10 Dimensions

Dimension	What it measures
Time-to-first-success	How fast a new user goes from landing to a working example
What-is-this clarity	Whether a stranger gets the value prop in 5 seconds
Visual proof	Gif/screenshot/video presence and position
Install simplicity	One-command vs. multi-step matrix
Runnable quick-start	Whether the example actually works
Documentation depth	README, docs/, external docs site
Examples	Number and quality of working examples
Community signals	CONTRIBUTING.md, CoC, templates, chat link
Trust signals	License, CI, recent commits, contributor count
Marketing signals	Tagline, homepage, social links, badges

Total: 100 points. Grades: A (90+), B (75–89), C (60–74), D (40–59), F (<40).

Input Parameters

Parameter	Required	Default	Description
`repo`	Yes	`.` (current dir)	Local path OR `https://github.com/owner/repo` URL
`tone`		`honest`	`brutal` / `honest` / `kind`
`output_dir`		`dx-roast/`	Where to write the 4 output files
`compare_to`			Optional second repo URL for benchmark comparison

Tone Reference

Tone	Voice	Best for
`brutal`	Battle-scarred CTO. Cutting, dryly funny. Punches the work, not the person.	Shareable Twitter post. Last-resort wake-up call.
`honest`	Senior engineer code review. Direct, kind, specific. (default)	Most cases. Internal sharing with the team.
`kind`	Encouraging mentor. Specific issues framed as opportunities.	Solo maintainers. First-time OSS authors.

Output

dx-roast/
├── roast.md           # 3–5 paragraph review in your tone
├── action-plan.md     # 5–10 prioritized fixes
├── ideal-readme.md    # Sketch of what your README could be
└── score.json         # Raw scores per dimension

The roast is built to be shareable. Tweet an excerpt. Paste in Slack. Forward to your co-founder.

Prompt Examples

Good:

Roast my repo. Tone: brutal.

Good:

DX audit on https://github.com/myorg/mytool. Compare to https://github.com/vercel/next.js.

Good:

Run dx-roaster on the current directory with tone=kind. Save outputs to ./pre-launch/.

Bad:

Tell me if my repo is good.

(Too vague. The skill needs at minimum a repo path; everything else has good defaults.)

Prompt Tips

Be brave. The brutal tone is funny, but it cuts deep. If you’re not ready to read what’s wrong with your repo, use honest or kind. Share the output. Roasts are designed to be shareable. A grade-D roast posted with the eventual fix is great open-build content. Run it before launch, not after. The action plan items are typically 5 minutes to 2 hours. Most are doable the night before Show HN. Compare to one repo, not five. The benchmark feature works best when you pick a single repo whose DX you envy.

Requirements

Claude Code, Gemini CLI, or GitHub Copilot
gh CLI (only needed if the input is a GitHub URL; not needed for local paths)
Read access to the target repo (local file system or public GitHub repo)

Catalog

Visual & Media

Content

Launch

GTM Intelligence

Outreach

Research

Developer Tools

Other

Install

How It Works

The 10 Dimensions

Input Parameters

Tone Reference

Output

Prompt Examples

Prompt Tips

Requirements

​Install

​How It Works

​The 10 Dimensions

​Input Parameters

​Tone Reference

​Output

​Prompt Examples

​Prompt Tips

​Requirements

Install

How It Works

The 10 Dimensions

Input Parameters

Tone Reference

Output

Prompt Examples

Prompt Tips

Requirements