Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Fetch PR review comments, display them, and optionally fix code issues raised by reviewers
.claude/skills/nudgebee-pr-comments-fetch-display-and-fix/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-13 | ✗→✓ | ▲ Improved | 25% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 16% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 24% | 0% |
Fetch and display comments from the current branch's pull request. Optional argument: $ARGUMENTS can be a PR number, fix to automatically address review comments, or fix 12345 to fix comments on a specific PR.
bash# Extract PR number from arguments (handles "fix", "fix 12345", "12345", or empty) PR_NUMBER=$(echo "$ARGUMENTS" | grep -o '[0-9]\+' | head -n 1) if [[ -z "$PR_NUMBER" ]]; then PR_NUMBER=$(gh pr view --json number -q .number) fi
If no PR is found, stop and tell the user: "No open PR found for the current branch. Usage: /pr-comments [PR_NUMBER]"
Run these commands to gather all comment types:
bashREPO=$(gh repo view --json nameWithOwner -q .nameWithOwner) # PR-level comments (issue comments) gh api /repos/${REPO}/issues/${PR_NUMBER}/comments # Code review comments (inline on diff) gh api /repos/${REPO}/pulls/${PR_NUMBER}/comments
Format and display all comments in a readable way:
- @author (date):
> comment body- @author file.go#line:
```diff
[diff_hunk]
```
> comment bodyRules:
$ARGUMENTS contains "fix")If the user invoked /pr-comments fix or asks to fix the comments:
bash git add <specific files> git commit --amend --no-edit git push --force-with-lease origin $(git branch --show-current)
git commit --amend -m "updated message" insteadbash # For inline review comments (code review comments): gh api /repos/${REPO}/pulls/${PR_NUMBER}/comments/${COMMENT_ID}/replies \ -f body="Fixed in $(git rev-parse --short HEAD)"
# For PR-level comments (issue comments): gh api /repos/${REPO}/issues/${PR_NUMBER}/comments \ -f body="Addressed in $(git rev-parse --short HEAD)"
Addressed N review comments:
Skipped M comments (questions/discussion):
Pushed commit: <sha>
| Path prefix | Validation command | |---|---| | api-server/services/ | cd api-server/services && make validate | | ticket-server/ | cd ticket-server && make validate | | collector-server/cloud-collector/ | cd collector-server/cloud-collector && make validate | | collector-server/k8s-collector/relay-server/ | cd collector-server/k8s-collector/relay-server && make validate | | collector-server/k8s-collector/app/ | cd collector-server/k8s-collector/app && make lint && make test | | llm/code-analysis/ | cd llm/code-analysis && make check | | llm/llm-server/ | cd llm/llm-server && make validate | | llm/rag-server/ | cd llm/rag-server && make lint && make test | | ml-k8s-server/ | cd ml-k8s-server && make lint && make test | | auto-pilot/ | cd auto-pilot && poetry run black --check . && poetry run flake8 . | | notifications-server/ | cd notifications-server && poetry run black --check . && poetry run flake8 . | | app/ | cd app && npm run lint2 |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-13 | fail→pass | 7,745 | 1,932 | -75% | 1 | 1 | 0% | 1,243 | 1,550 | +25% | 0 | 0 | — |
case-02 | fail→fail | 6,528 | 6,363 | -3% | 1 | 1 | 0% | 976 | 1,505 | +54% | 0 | 0 | — |
case-03 | fail→fail | 2,655 | 4,850 | +83% | 1 | 1 | 0% | 357 | 1,530 | +329% | 0 | 0 | — |
case-01 | fail→fail | 13,473 | 4,920 | -63% | 1 | 1 | 0% | 2,395 | 1,530 | -36% | 0 | 0 | — |
case-04 | pass→fail | 7,788 | 5,166 | -34% | 1 | 1 | 0% | 1,269 | 1,475 | +16% | 0 | 0 | — |
case-05 | pass→pass | 10,563 | 6,288 | -40% | 1 | 1 | 0% | 1,768 | 2,332 | +32% | 0 | 0 | — |
case-06 | fail→fail | 3,421 | 4,098 | +20% | 1 | 1 | 0% | 448 | 1,413 | +215% | 0 | 0 | — |
case-07 | pass→pass | 9,458 | 3,865 | -59% | 1 | 1 | 0% | 1,612 | 1,893 | +17% | 0 | 0 | — |
case-08 | pass→pass | 3,402 | 3,047 | -10% | 1 | 1 | 0% | 537 | 1,797 | +235% | 0 | 0 | — |
case-09 | pass→pass | 4,544 | 2,482 | -45% | 1 | 1 | 0% | 770 | 1,677 | +118% | 0 | 0 | — |
case-10 | fail→pass | 6,868 | 1,612 | -77% | 1 | 1 | 0% | 1,112 | 1,484 | +33% | 0 | 0 | — |
case-11 | fail→pass | 8,166 | 1,709 | -79% | 1 | 1 | 0% | 1,369 | 1,567 | +14% | 0 | 0 | — |
case-12 | fail→pass | 7,977 | 1,211 | -85% | 1 | 1 | 0% | 1,267 | 1,472 | +16% | 0 | 0 | — |
case-14 | pass→pass | 7,585 | 1,451 | -81% | 1 | 1 | 0% | 1,243 | 1,540 | +24% | 0 | 0 | — |
case-15 | fail→pass | 12,897 | 1,263 | -90% | 1 | 1 | 0% | 1,186 | 1,468 | +24% | 0 | 0 | — |
case-16 | fail→pass | 7,208 | 1,146 | -84% | 1 | 1 | 0% | 1,098 | 1,410 | +28% | 0 | 0 | — |
case-17 | fail→pass | 8,846 | 1,403 | -84% | 1 | 1 | 0% | 1,421 | 1,458 | +3% | 0 | 0 | — |
case-18 | fail→pass | 9,937 | 1,077 | -89% | 1 | 1 | 0% | 1,434 | 1,453 | +1% | 0 | 0 | — |
case-19 | fail→pass | 22,478 | 1,506 | -93% | 1 | 1 | 0% | 2,448 | 1,450 | -41% | 0 | 0 | — |
case-20 | fail→pass | 8,866 | 1,899 | -79% | 1 | 1 | 0% | 1,497 | 1,587 | +6% | 0 | 0 | — |
case-21 | pass→pass | 7,332 | 1,944 | -73% | 1 | 1 | 0% | 1,311 | 1,615 | +23% | 0 | 0 | — |
case-22 | pass→pass | 7,812 | 2,144 | -73% | 1 | 1 | 0% | 1,299 | 1,592 | +23% | 0 | 0 | — |
case-23 | pass→pass | 3,984 | 1,803 | -55% | 1 | 1 | 0% | 651 | 1,569 | +141% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted, and 18 counted toward the lift figure. The other 5 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +39 percentage points is the difference between those two pass rates over the 18 comparable cases. 3 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.