pr-buddy

The problem

Agentic PR review tools are fast. Too fast. Teams accumulate cognitive debt — reviewers click "OK-APPLY" without ever building a mental model of the changes. Velocity increases. Comprehension does not.

Each rubber-stamped PR leaves your team less equipped to reason about the next one. Incident response slows down. Onboarding degrades. Ownership diffuses.

How it works

/pr-buddy 847 │ ├── Pre-phase fetch PR + diff (silent) │ ├── Phase 1 Socratic walkthrough │ │ │ ├── Station 1 · Goal │ ├── Station 2 · Architecture │ ├── Station 3 · Decisions │ ├── Station 4 · Risk │ └── Station 5 · Ownership ← gate │ ├── Phase 2 pr-review-toolkit (anchored to Phase 1) │ └── Summary 3-bullet recap → paste into PR comment

Install

claude skill add manuartero/pr-buddy

Requirements: gh CLI (authenticated) · pr-review-toolkit for Phase 2

Usage

/pr-buddy <pr_number_or_branch_or_url>

# examples
/pr-buddy 847
/pr-buddy feat/rate-limiting
/pr-buddy https://github.com/org/repo/pull/847

The 5 stations

Each station follows the same pattern: you answer first, the skill reveals its reading second.

Goal

"What problem do you think this PR is solving?"

Architecture

"Which is the entry point? Walk me through your mental model."

Decisions

"The author chose X. What would you have done differently?"

Risk

"If you had to name one thing that could go wrong here, what would it be?"

Ownership

"Could you explain this PR to a teammate? What are you still unsure about?"

← Phase 2 does not start until you confirm readiness here.