Review a PR Locally (ADO + GitHub)

Check out a pull request into an isolated git worktree — never disturbing the user's current checkout — surface its changes in the editor, run an interactive review where the human drives, and post only the comments the human asks for. All identifiers are derived from the PR URL, so the skill works against any ADO or GitHub repository.

Review Philosophy — high-signal, human-owned

This is the core of the skill. Violating it defeats the purpose.

• The human drives the review. The AI surfaces the parts worth attention and answers the human's questions. It never decides on its own to flood the PR with comments.
• Comments are posted only on explicit human request, in one of two ways:
  1. Human-dictated: the human gives exact location + text (e.g. post comment on line 53 of src/foo.ts - this returns before the lock is released). Post it verbatim.
  2. AI-drafted on request: the human asks the AI to write a comment for a specific spot (e.g. add a comment on the null check in bar.ts). Draft it terse and high-signal, show it, and post only after the human approves.
• When surfacing "important parts," comment-worthy means: correctness bugs, logic errors, security issues, breaking API changes, missing edge cases, data loss/corruption, race conditions, broken error handling. Ignore style, formatting, and naming.
• Never post praise, summaries, restatements of what the code does, or style nits — in either mode.
• Every posted comment cites a concrete file:line and is 1–3 sentences.

Step 1 — Parse the PR URL

Detect the platform and extract every identifier from the URL — nothing is hardcoded.