Review Alerts

Review Alerts

Show a prioritized summary of PRs waiting for the caller's review. Never starts an actual review.

Prerequisites

This skill expects:

• gh CLI installed and authenticated
• jq available on PATH

Step 1: Validate Inputs

This skill requires two parameters: repo (GitHub owner/name) and user (GitHub login).

If either is missing, ask the user:

I need two things to check your reviews:
- **repo**: GitHub repository (e.g., `owner/repo`)
- **user**: Your GitHub username (e.g., `octocat`)

Which repo and user should I check?

And stop until the user provides them. Do NOT auto-detect from git context or gh api user.

Step 2: Discover PRs

Run the bundled discovery script with explicit flags:

PR_JSON=$("/find-my-reviews.sh" --detailed --repo "$REPO" --user "$USER")

Where ` is the directory containing this SKILL.md` file.

If the script returns an empty array ([]), respond with:

Nothing to review right now. Enjoy your day!

And stop.

Step 3: Assess Priority and Review Complexity

Priority

For each PR in the JSON, assign a priority level:

High (any of):

• Label contains: urgent, hotfix, blocking, critical, P0, P1, high-priority, expedite
• Base branch matches: release/*, hotfix/*, production
• Age 7+ days

Medium (any of):

• Age 3-7 days
• Moderate size with activity

Low:

• Created less than 3 days ago, no urgency signals

Sort: High first, then Medium, then Low. Within same priority, oldest first.

Review Complexity

For each PR, assess review complexity:

Low: changed_files 15 OR (additions + deletions) > 500 OR touches auth/crypto/security paths Medium: everything else

Security-sensitive path patterns: auth/, crypto/, keychain/, Token, Password, Secret, security/, credential

Step 4: Format the Report as ASCII Table

Generate a description for each PR: first sentence of the PR body, or "No description" if empty. Truncate to ~60 characters.

Use box-drawing characters for the table. The table IS the output - no per-PR detail sections.

## Pending Reviews

You have **N** PR(s) waiting for your review.

┌───┬─────────────────────────┬──────────┬──────┬────────────────┬──────────┬──────────────────────────────────┬────────────┐
│ # │ PR                      │ Author   │ Age  │ Size           │ Priority │ Description                      │ Complexity │
├───┼─────────────────────────┼──────────┼──────┼────────────────┼──────────┼──────────────────────────────────┼────────────┤
│ 1 │ #123 - Fix auth flow    │ @alice   │ 5d   │ +45 -12 (3f)   │ High     │ Fixes OAuth redirect loop        │ Medium     │
│ 2 │ #456 - Add dark mode    │ @bob     │ 2d   │ +200 -50 (8f)  │ Low      │ Dark mode support for settings   │ Medium     │
└───┴─────────────────────────┴──────────┴──────┴────────────────┴──────────┴──────────────────────────────────┴────────────┘

Rules:

• Age format: "<1d", "1d", "5d", "1w", "2w" etc.
• Size format: +N -N (Xf) where f = files
• PR column: #number - Title (truncate title to fit ~25 chars). Include the full GitHub link as a markdown link.
• Description: first sentence of PR body, truncated to ~60 chars. "No description" if empty.
• Complexity column: Low, Medium, or High

Step 5: Review History Bar Chart

Fetch the user's review activity for the last 5 days using the GitHub search API:

# For each of the last 5 days, count PRs reviewed
gh api "search/issues?q=repo:$REPO+type:pr+reviewed-by:$USER+updated:YYYY-MM-DD..YYYY-MM-DD" --jq '.total_count'

Render a horizontal bar chart using block characters, scaled to a max width of 10 characters. Find the largest count across all 5 days — that day gets 10 blocks, and the others scale proportionally (round to nearest, minimum 1 block for non-zero counts):

Reviews completed (last 5 days):

Mar 12  ████████████   6
Mar 11  ███████████    5
Mar 10  ████           2
Mar 09                 0
Mar 08  ██████████     3
                       ─────
                       Total: 16

Use █ (full block) characters. If a day has 0 reviews, show no blocks and just the 0. Right-align the count numbers after the bar area. Show the total at the bottom.

If the API call fails or returns errors, skip the chart silently - it's supplementary.

Step 6: Stop

After outputting the report and chart, STOP. Do not create worktrees, do not start reviewing code, do not spawn subagents. This skill is report-only.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: boundsj
• Source: boundsj/agent-skills
• License: MIT

Install and usage instructions live in the source repository linked above.