roast-rematchlisted
Install: claude install-skill vivekbhange-hub/repo-roast-skill
# roast-rematch: the comeback episode
The best part of a roast is the comeback. This skill re-convenes the same panel with the same
rubric on the fixed repo, then lines the two verdicts up side by side so the improvement is
measured, not claimed. The before and after card is the one people post.
## Inputs
- The previous roast: the path the user gives, else the newest file in `roast/history/`, else the
current `roast/roast.json` (which will be archived when the new roast is written).
- Mode flag: use the previous roast's `mode` unless the user asks for another. Changing tone
between rounds is fine; changing the rubric is not.
## Outputs
- `roast/roast.json`, `roast/ROAST.md`: the new roast (the old one is archived to
`roast/history/` by verdict.py).
- `roast/REMATCH.md`: the delta report.
- `roast/card.json` and `roast/card.svg`: the card with before and after scores.
## Steps
1. **Find the baseline.** Locate the previous roast as above. Read its `date`, `mode`, `overall`
and per-judge scores. If there is no previous roast, say a rematch needs a first match and run
`roast-repo` instead.
2. **Note what changed since.** `git log --oneline --since=<previous date>` and
`git diff --stat` for uncommitted work. If nothing changed, say so and ask whether to re-roast
anyway (scores may still move a little because judges read fresh).
3. **Run the full roast again.** Follow `roast-repo` steps 2 to 6 exactly (facts, the six judges in
parallel, save panel file