pr-review-loop

Repeats structured code review and root-cause fixes on branch diffs until findings converge.

1|Updated Jan 31, 2026
One-click install
npx skills add https://github.com/tktcorporation/cli-sim-game-escape --skill pr-review-loop-tktcorporation
Or copy as Structured Prompt for Agent
Please help me install this Agent Skill.
Skill: pr-review-loop
Source: https://github.com/tktcorporation/cli-sim-game-escape/tree/main/.claude/skills/pr-review-loop
Command: npx skills add https://github.com/tktcorporation/cli-sim-game-escape --skill pr-review-loop-tktcorporation

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? One-shot code reviews often leave residual issues, and fixing findings one by one with local patches creates whack-a-mole cycles. This Skill automates repeated review-and-fix rounds on your local branch diff before PR creation, clustering findings by root cause so each round eliminates whole classes of problems instead of individual symptoms. ## Core Features & Use Cases - Iterative Review Loop: Runs a dedicated sub-agent per round to review the current diff with a four-lens structured review, keeping the main session context lean. - Root-Cause Clustering: Groups each round's findings by shared design causes (responsibility placement, data normalization, contract ambiguity) and fixes clusters with upstream fixes rather than per-finding patches. - Convergence Detection: Stops when findings reach zero or when the underlying root causes stop changing between rounds, with a configurable round limit and user confirmation on exhaustion. - Use Case: Before opening a PR, ask for repeated review and fixing until no findings remain; the Skill loops review, diagnosis, and fix rounds, then reports a per-round summary of findings, root causes, and resolutions. ## Quick Start Review my current branch diff against main and keep fixing the findings until the review converges with nothing left.

Frequently Asked Questions about pr-review-loop

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I run repeated code reviews until all findings are fixed?

Invoke the loop with a request like 'review and fix until findings converge' on your local branch. It diffs against origin/main, reviews each round in a dedicated sub-agent, fixes root causes, and repeats until zero findings or semantic convergence.

How does automated PR review avoid fixing the same issue repeatedly?

It clusters each round's findings by shared root causes such as responsibility placement or contract ambiguity, then applies one upstream fix per cluster. Convergence is judged by comparing diagnosed root causes across rounds, not raw finding text.

When should I use a review loop instead of a single deep review?

Use the loop when you explicitly want repeated review-and-fix cycles before PR creation. A single deep review suffices for one-pass feedback, and a single clear bug fix only needs a root-cause fix without iteration.

What happens when the review loop reaches its round limit?

By default the loop stops after five rounds and asks the user whether to continue, reporting the finding count trend and why convergence was not reached. It never silently keeps looping or silently gives up.

Does this replace pre-PR checks like codex review?

No. Convergence of this loop does not exempt the separate pre-PR gate such as running codex review twice before gh pr create. The loop provides structured four-lens review while the gate is a lightweight general check, so both apply.