| name | reply-to-pr-thread |
| description | Use when the user asks to reply / comment on a CodeRabbit (or any) PR review thread, OR when posting a reasoning/explanation reply tagging @coderabbitai for a thread you've classified as invalid. Wraps `gh api .../pulls/N/comments/<id>/replies`. Body is read from stdin. Auto-approved globally. |
Reply to a GitHub PR review thread
Wraps the GitHub POST /repos/{owner}/{repo}/pulls/{pr}/comments/{id}/replies endpoint. Works for any review thread on any PR, but the canonical use case is replying to CodeRabbit threads — never resolve them, always reply with rationale (per the user's global memory feedback_coderabbit_threads.md).
How to run
echo "@coderabbitai <reason>" | \
bash ~/.claude/skills/reply-to-pr-thread/scripts/reply-to-pr-thread.sh \
https://github.com/OWNER/REPO/pull/<PR>#discussion_r<COMMENT_ID>
bash ~/.claude/skills/reply-to-pr-thread/scripts/reply-to-pr-thread.sh \
--comment-id <COMMENT_ID> --pr <PR> [--repo OWNER/REPO] <<'EOF'
@coderabbitai
multi-line
reasoning here
EOF
- Body is always read from stdin. No
--body / --body-file. Pipe text in or use a heredoc.
- On success, the script prints the new reply's
html_url. Report that URL back to the user.
Steps
- Identify the comment URL or comment ID. From the
fetch-coderabbit-threads output, that's the URL printed in Comment 1 — coderabbitai. The integer after discussion_r is the comment ID.
- Compose the reply body. For a CodeRabbit-invalid case: lead with
@coderabbitai, then 1–3 sentences explaining why the comment doesn't apply (cite the relevant CLAUDE.md rule, the existing test, or the commit hash where it was already addressed).
- Pipe the body into the script. Capture the printed URL.
- Never call any "resolve" mutation (
resolveReviewThread GraphQL or its UI equivalent). This script is the only allowed channel for closing the loop on a CodeRabbit thread you've classified as invalid.
Why use this instead of gh api directly
- Auto-approved — no permission prompt.
- Parses discussion URLs so you don't have to hand-extract owner/repo/PR/comment-id.
- Uses
jq to build the JSON payload, so multi-line bodies with quotes / backticks / Markdown round-trip safely.
- Single output line (the reply URL) makes it easy to chain.
If the script is missing a flag you need, extend the script rather than reaching for gh directly.
