vivekbhange-hub/repo-roast-skill

Get your repo roasted by a panel of six brutally honest senior engineers. Every jab comes with evidence and a fix.
Example card for SnackStack, a made-up repo. Your card is generated from your own roast.
Code reviews are either too polite to be useful or too vague to act on. "Consider improving test coverage" changes nothing. "There is no CI, so the test suite runs whenever someone remembers, which is never" gets fixed the same afternoon, especially when it comes with the exact workflow file to add.
This Claude Code plugin puts your repository in front of six judges, each a subagent with its own personality and its own scoring rubric: The Paranoid (security), The Stopwatch (performance), The New Hire (first-day developer), The Greybeard (twenty-year maintainer), The User (product and UX) and The On-Call (ops and reliability). They read your code in parallel and come back with jokes, scores and fixes.
The catch, and the point: no evidence, no roast. Every jab must cite a real file:line, a command and its output, or a measured fact about the repo. A script checks every citation against your files and throws out anything that does not hold up, in public, at the bottom of the report. Every score is the sum of a five-part rubric. Tone goes from --gentle to --savage; the evidence bar never moves.
What you get: roast/ROAST.md with an overall score, each judge's roast, the top 10 fixes ranked by impact for the effort and the one thing to fix today, plus a 1200x630 SVG scorecard and a README badge you can post. Fix things, run a rematch, and the card shows before and after.
## What's inside
### Skills
| Skill | What it does |
| roast-repo | The main event. Gathers facts, runs the six judges in parallel, validates every citation, writes roast/ROAST.md, the card and the badge. |
| roast-card | Renders the shareable 1200x630 SVG scorecard and a "roast score 7.4/10" badge, with before and after support. |
| roast-readme | Roasts only the README and the first five minutes a stranger spends with your project, then rewrites the top (saved only after you approve the diff). |
| roast-pr | Roasts a diff or pull request, citing new-side line numbers. Drafts polite review comments; posts nothing without a yes. |
| roast-fix | Turns the roast into roast/FIXPLAN.md and, on approval, fixes quick wins one at a time with your tests as the referee. Never commits. |
| roast-rematch | Re-roasts after fixes and shows the score change per judge, dropped and new charges, and a before and after card. |
### Judges (agents, all read-only)
| Agent | Persona | Scores |
| judge-security | The Paranoid | secrets, injection and unsafe calls, dependencies, auth, safe defaults |
| judge-performance | The Stopwatch | data access (N+1), hot-path algorithms, blocking IO and timeouts, payload size, resource use |
| judge-newcomer | The New Hire | first impression, setup path, run and verify, naming, help on hand |
| judge-maintainer | The Greybeard | tests, CI, structure, error handling, dead code and hygiene |
| judge-product | The User | problem clarity, promises kept, interface ergonomics, error messages, accessibility and polish |