How do I install the Watch Pr skill?

Run npx skillmds@latest add sherifabdlnaby/watch-pr in your terminal (requires Node.js), paste this page's agent-chat prompt into Claude, Cursor, or any MCP-connected agent, or download the SKILL.md file and copy it into your agent's skills directory.

What does the Watch Pr skill do?

Watch a PR until CI is green and its reviews are answered. It is listed under Coding & Dev Tools on SkillMD.

Is Watch Pr safe to use?

This skill has not completed SkillMD's automated safety review yet. SkillMD never runs a skill's scripts for you; review the SKILL.md before installing.

Which AI agents work with Watch Pr?

This skill is tagged as working with Claude Code, Claude.ai, OpenAI Codex. SKILL.md is an open format, so most agents that read a skills directory can load it too.

Is Watch Pr free to use?

Yes. Installing skills from SkillMD is free. This skill is licensed under MIT.

Who published Watch Pr?

sherifabdlnaby (@sherifabdlnaby) published this skill. Their other Agent Skills are listed on their SkillMD profile.

Watch Pr by sherifabdlnaby — Agent Skill

Name: Watch Pr
Author: sherifabdlnaby

File contents

watch-pr

Load the git skill, then read its watch and review-responses references before acting. <git-skill-dir> below is the base directory the harness printed when the git skill loaded.

Goal: get the PR to green with every review answered, without blocking your turn. Fix CI failures, address bot reviews, answer human comments per review-responses, and escalate only what needs my input while you keep making progress on everything else. Block waiting on me only once CI is green and every bot review is addressed.

The sibling skills narrow this: watch-pr-ci (CI only) and watch-pr-comments (reviews only).

Target PR

$ARGUMENTS may name the PR (number or URL). If empty, resolve the PR for the current branch:

gh pr view --json number,url,headRefName,state,isDraft

Start

Draft PR: run flick once, alongside the watch, not before it. Chase beyond that (hold it open, flick again) only if I told you this repo has a review bot; me naming the bot counts.
Existing review threads: unresolved threads already on the PR are not news to the watch. Fetch them per review-responses Batching and address them first.

The watch

python3 <git-skill-dir>/scripts/pr-watch.py watch --pr <N> --repo <OWNER/REPO> --watcher <id> --until quiet --max-total <s>

Run it through the cheap sub-agent the watch reference describes. React per the verdict: EVENT means act and relaunch, STALE means one ⚠️ line for me and relaunch, QUIET means relaunch, DONE means digest and stop.

Time Budget

$ARGUMENTS may include a duration (30m, 1h, 2h, …) or forever / indefinitely.

Duration -> --max-total <seconds>. Default 30m when neither a duration nor forever is given.
forever -> no --max-total, and for this skill --until closed, so settling does not end it; the siblings keep their own --until.

Human comments

Human reviews and comments are answered per review-responses: fix what is clearly right, push back on what is not, and bring the real trade-offs to me instead of deciding them. With humans in $ARGUMENTS, decide those too and post the reply as Agent Decided.

---name: watch-prdescription: Watch a PR until CI is green and its reviews are answered.license: MIT---# watch-prLoad the **git** skill, then read its watch and review-responses references before acting.`<git-skill-dir>` below is the base directory the harness printed when the git skill loaded.Goal: get the PR to green with every review answered, without blocking your turn. Fix CI failures,address bot reviews, answer human comments per review-responses, and escalate only what needs myinput while you keep making progress on everything else. Block waiting on me only once CI is greenand every bot review is addressed.The sibling skills narrow this: `watch-pr-ci` (CI only) and `watch-pr-comments` (reviews only).## Target PR`$ARGUMENTS` may name the PR (number or URL). If empty, resolve the PR for the current branch:```gh pr view --json number,url,headRefName,state,isDraft```## Start1. **Draft PR:** run `flick` once, alongside the watch, not before it. Chase beyond that (hold it open, flick again) only if I told you this repo has a review bot; me naming the bot counts.2. **Existing review threads:** unresolved threads already on the PR are not news to the watch. Fetch them per review-responses [Batching](../../git/references/review-responses.md#batching) and address them first.## The watch```python3 <git-skill-dir>/scripts/pr-watch.py watch --pr <N> --repo <OWNER/REPO> --watcher <id> --until quiet --max-total <s>```Run it through the cheap sub-agent the watch reference describes. React per the verdict: `EVENT`means act and relaunch, `STALE` means one ⚠️ line for me and relaunch, `QUIET` means relaunch, `DONE`means digest and stop.## Time Budget`$ARGUMENTS` may include a duration (`30m`, `1h`, `2h`, …) or `forever` / `indefinitely`.- **Duration** -> `--max-total <seconds>`. Default **30m** when neither a duration nor `forever` is given.- **`forever`** -> no `--max-total`, and for this skill `--until closed`, so settling does not end it; the siblings keep their own `--until`.## Human commentsHuman reviews and comments are answered per review-responses: fix what is clearly right, push backon what is not, and bring the real trade-offs to me instead of deciding them. With `humans` in`$ARGUMENTS`, decide those too and post the reply as Agent Decided.

Frequently asked questions

Watch Pr