Make every agent-written comment easy to spot, short and safe to publish.
Steps
-
Write the reply. Lead with the conclusion. Cite the code line, check, test or reproduction that supports it. Mark what you verified and what is your opinion.
-
Put this note at the top, with the reply below it:
> [!NOTE] > **<model> responding on behalf of <user>** Reply text starts here.Use the model name the harness reports. If it reports none, use the harness name. Use the user's name or GitHub login.
-
Scan the full comment, including quotes and attachments, for secrets, tokens, private paths and personal data. Remove them.
-
Preview the final Markdown. Confirm the note and the reply render as two separate blocks.
-
Post the comment.
-
Confirm GitHub returned the comment or review URL. Print it.
-
Resolve a review thread only after its reply posted.
Stop when
- GitHub returned a URL for each comment you meant to post.
- A post fails twice. Report the error and do not resolve the thread.
Rules
- Use the note on top-level comments, inline review comments, replies to bots and people, and dismissal reasons.
- Do not use it in PR descriptions, commit messages or text the user asked to post word for word.
- One attribution only. No signature, footer, heading or horizontal rule.
- No thanks, filler or status updates such as "still working on it".
- Never guess the model name.
