MR Comment Handler Skill

MR Comment Handler

Find the open MR for the current branch and address its review threads using glab. Run all glab commands with elevated network access.

Prerequisites

Ensure glab auth status succeeds (via glab auth login or GITLAB_TOKEN).
Ensure glab is at least v1.80.4.
When sandboxing blocks network calls, rerun with sandbox_permissions=require_escalated.
Sanity check auth up front:

glab auth status

1) Resolve the MR for the current branch

Do a quick check so we know which MR we are about to operate on:

branch="$(git rev-parse --abbrev-ref HEAD)"
glab mr view "$branch" --output json

If this fails, the fetch script below will still try to locate the MR by source_branch.

2) Fetch unresolved discussions to `/tmp`

Use the local script to fetch MR discussions via glab api. This filters out bot/system-only threads and returns unresolved discussions when --open-comments is set.

skill_dir="<path-to-skill>"
branch="$(git rev-parse --abbrev-ref HEAD)"
safe_branch="${branch//\//_}"
out="/tmp/${safe_branch}_mr_open_discussions.json"
python "$skill_dir/scripts/fetch_comments.py" --open-comments --output "$out"

If you want the full payload (including resolved discussions), drop --open-comments.

3) Summarize, triage, and ask once

Load the JSON and number each unresolved discussion.
Start with a compact summary list instead of dumping full threads.
Sort by: unresolved first (already filtered), then most recently updated.
When possible, group or label by file path to reduce context switching.
In the summary list, show: number, discussion id, author, file:line (if present), and a one-line summary.
Ask for a batch selection in one shot. Accept: 1,3,5-7, all, none, or top N.
If the user does not choose, suggest a small default set (for example top 3) with a short rationale.
Only after selection, show the full thread and code context for the selected numbers.
When showing code context, include 3 lines before and after and clearly mark the referenced line(s).

4) Implement fixes for the selected discussions

Apply focused fixes that address the selected threads.
Run the most relevant tests or checks you can in-repo.
Report back with: what changed, which discussion numbers were addressed, and any follow-ups.

Notes:

If glab hits auth or rate issues, prompt the user to run glab auth login or re-export GITLAB_TOKEN, then retry.
If no open MR is found for the branch, say so clearly and ask for the MR URL or IID.
Do not prompt “address or skip?” one comment at a time unless the user explicitly asks for that mode.

Agent Skills: MR Comment Handler

Install this agent skill to your local

Skill Files