| name | review-pr-ado |
| description | Code review an Azure DevOps (ADO) pull request. Use whenever the user asks to review a PR, code-review a pull request, critique/check an ADO PR, runs /review-pr-ado, or gives a PR number with review intent. ADO repos are NOT on GitHub — there is no `gh`; this skill drives review through the bundled `ado-pr-review` MCP server (ado_pr_show / ado_pr_changed_files / ado_pr_threads_list / ado_pr_list_active / ado_pr_thread_post) with local `git` for diff bodies. It runs a multi-lens parallel review via the bundled ado-review-* agents, adversarially scores each finding 0-100, keeps only high-confidence (>=80) issues, and — dry-run by default — posts a review comment only when --post is passed. |
review-pr-ado (Claude Code plugin)
Disciplined PR review for Azure DevOps repos. The capability layer is the bundled MCP server
ado-pr-review; the workflow is the shared playbook. This skill wires them together and dispatches
the bundled lens agents in parallel.
MCP tools (from the bundled server)
ado_pr_show — PR metadata (status, draft, branches, merge commits, project, repo id). Start here.ado_pr_changed_files — files changed by the PR.ado_pr_threads_list — existing comment threads (eligibility / no double-post).ado_pr_list_active — active PRs (concurrent-PR / merge-conflict check).ado_pr_thread_post — post the review comment. Outward-facing — only on --post.
If the MCP server isn't loaded, every call has a raw az equivalent in
references/ado-commands.md. Diff bodies come from local git (see references) — the MCP
server returns changed-file lists, not diff text.
Arguments
- PR id (accept
1234, #1234, or an ADO PR URL — extract the number).
--post → publish the comment. Default is dry-run: present findings, post nothing.
Workflow
Make a todo list from these steps.
- Eligibility —
ado_pr_show. Skip (say why) if abandoned/completed, draft, automated/trivial,
or already reviewed (a prior ### Code review thread in ado_pr_threads_list). Else stop.
- Gather — read the repo-root
CLAUDE.md/AGENTS.md (+ any in touched dirs); ado_pr_changed_files
to scope lenses; fetch the diff via local git using lastMergeSourceCommit/lastMergeTargetCommit
from ado_pr_show (see references/ado-commands.md); write a 2-3 sentence summary.
- Multi-lens review — dispatch the bundled agents in parallel, each with the diff inline,
scaling to PR size. Agents:
ado-review-guidance, ado-review-bugs, ado-review-history,
ado-review-comments, ado-review-tenant. Each returns issues + a one-line reason.
- Score — for each distinct finding, dispatch
ado-review-scorer (cheap/skeptical) to score it
0-100 with the rubric in references/review-lenses.md. De-duplicate first.
- Filter — keep only ≥ 80. Nothing clears? Report "no issues" plainly.
- Verify survivors — re-confirm each high-impact claim with a real tool/command before posting.
- Re-check eligibility —
ado_pr_show + ado_pr_threads_list again.
- Deliver — dry-run: present findings (format below).
--post: ado_pr_thread_post, then read
back via ado_pr_threads_list to confirm.
Full lens checklists, the multi-tenant rules, the scoring rubric, and the known ADO false
positives (e.g. pr: YAML trigger is a no-op on Azure Repos) live in references/review-lenses.md.
ADO/az/git command cookbook + citation URL format: references/ado-commands.md.
Output format
Brief, no emojis (except footer). Cite each finding with an ADO file link at the head commit; link
any referenced PR by its ADO URL.
### Code review
Found N issues:
1. <brief description> (<reason>)
<ADO file link with path + line range at the head commit SHA>
🤖 Generated with ado-pr-review
No issues clear the filter:
### Code review
No issues found. Checked for bugs, CLAUDE.md/AGENTS.md compliance, tenant isolation, and ADO/CI pitfalls.
🤖 Generated with ado-pr-review
