Merged PRs Summary

Summarize completed (merged) pull requests across the user's Azure DevOps server within a time window.

Arguments

$ARGUMENTS — optional time window. Examples: 24h, 48h, 7d, 3d. Defaults to 24h when empty.

Parse it as <number><unit> where unit is h (hours) or d (days). Convert days to hours (1d = 24h).

Azure DevOps server

• Base URL: https://devops.example.com
• Auth: Windows Integrated (curl --negotiate -u :)
• Projects: discover at runtime via /DefaultCollection/_apis/projects?api-version=7.0. e.g. MyProject (primary) and Sandbox.

Steps

1. Parse the window. Determine the cutoff datetime as now - <window>. If $ARGUMENTS is empty or unparseable, fall back to 24 hours.
2. List projects.

   curl -s -L --negotiate -u : "https://devops.example.com/DefaultCollection/_apis/projects?api-version=7.0"

   Extract project name values from the JSON value array.
3. Fetch completed PRs per project. For each project name:

   curl -s -L --negotiate -u : "https://devops.example.com/DefaultCollection/<ProjectName>/_apis/git/pullrequests?searchCriteria.status=completed&api-version=7.0&\$top=200"

   The API returns most-recent-first. If the oldest PR returned is still newer than the cutoff, increase $top and re-fetch (some windows like 7d will need 500+).
4. Filter by closedDate — keep PRs where closedDate >= cutoff. Note: closedDate is ISO-8601 with fractional seconds and a Z suffix; parse robustly.
5. Group by repository (p['repository']['name']). Within each repo, list PRs as PR <id> | <author display name> | <title>.
6. Render the summary in this shape:
   • Lead line: total PR count across all projects + the window.
   • Sections by repo, sorted by PR count descending. Use tiered headers (Heavy / Moderate / Light / Singletons) when the spread justifies it; otherwise just one flat list.
   • Within each repo entry, group multiple PRs by the same author when natural ("Jane Doe refactoring (X, Y, Z)") rather than listing each as a bare bullet.
   • Skip projects with zero PRs in window, but mention them briefly ("Sandbox had none") so the user knows they were checked.
   • End with a short Themes paragraph: cross-cutting patterns (e.g. "solution-structure cleanup dominates", "the BI team is mid-migration to a new scheduler"), or "No obvious cross-repo incident" if nothing stands out.

Python one-liner for filtering/grouping

The data is easier to handle in Python than jq on Windows. Pattern that works:

curl -s ... | python -c "
import sys, json
from datetime import datetime, timedelta, timezone
from collections import defaultdict
d = json.load(sys.stdin)
cutoff = datetime.now(timezone.utc) - timedelta(hours=<N>)
prs = [p for p in d['value']
       if datetime.fromisoformat(p['closedDate'].replace('Z','+00:00').split('.')[0]+'+00:00') >= cutoff]
by_repo = defaultdict(list)
for p in prs: by_repo[p['repository']['name']].append(p)
for repo, items in sorted(by_repo.items(), key=lambda x: -len(x[1])):
    print(f'=== {repo} ({len(items)}) ===')
    for p in items:
        print(f\"  PR {p['pullRequestId']} | {p['createdBy']['displayName']} | {p['title']}\")
"

Notes

• Don't fetch PR descriptions/diffs by default — title + author is enough for the overview. Only drill into a specific PR if the user asks "tell me more about PR X".
• The PR title is usually the merge commit subject (DevOps strips the Merged PR <id>: prefix on the web UI but the API title is clean).
• Windows curl handles the URL escaping fine, but the $top query param must be escaped as \$top in bash to avoid shell expansion.
• If the user is on a feature branch and only cares about one repo, this skill is overkill — point them at git log --since=... instead.