Review Requests

Batch-open outstanding GitHub review requests in pair-review with AI analysis.

Workflow

Phase 1: Discover pair-review server

Call mcp__pair-review__get_server_info to get the pair-review web UI URL. If the server is not running, inform the user and stop.

Phase 2: Find pending review requests

Use user-review-requested:@me (not review-requested) to find only PRs where the user was directly requested as a reviewer, excluding team-based review requests. The gh search prs CLI does not support this qualifier, so use the search API directly:

SINCE=$(date -v-7d +%Y-%m-%d 2>/dev/null || date -d '7 days ago' +%Y-%m-%d)
gh api "search/issues?q=is:pr+is:open+user-review-requested:@me+updated:>=${SINCE}&per_page=30" \
  --jq '.items[] | {number, title, html_url, repo: (.repository_url | split("/")[-2:] | join("/"))}'

Each result provides number, title, html_url, and repo (as owner/repo).

Phase 3: Open each PR in pair-review with auto-analysis

Limit: Open at most 5 PRs. If more than 5 are found, open only the first 5 (most recently updated) and report how many were skipped.

For each PR found (up to the limit), open it in the browser with the ?analyze=true query parameter, which automatically starts AI analysis when the page loads:

open "{server_url}/pr/{owner}/{repo}/{number}?analyze=true"

where {owner}/{repo} comes from the repo field split on /.

No MCP calls per PR are needed — the ?analyze=true parameter handles triggering analysis.

Phase 4: Report

Summarize what was done:

• How many PRs were found with pending review
• List each PR opened in pair-review with title and URL
• If no pending review requests exist, say so