harden-pr

Iteratively review and repair GitHub pull requests using model-diverse independent review passes until convergence.

1|Updated Apr 15, 2026
One-click install
npx skills add https://github.com/pnewsam/skills --skill harden-pr-pnewsam
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: harden-pr
Source: https://github.com/pnewsam/skills/tree/main/archive/ship-pr-merge-evicted/harden-pr
Command: npx skills add https://github.com/pnewsam/skills --skill harden-pr-pnewsam

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve? Open pull requests often accumulate review feedback, hidden defects, and unresolved threads that require repeated review-and-fix cycles before they are safe to merge. This Skill automates that hardening loop with independent, model-diverse reviews and traceable repairs while preventing scope creep and unverified merge-ready claims. ## Core Features & Use Cases - Bounded review-repair cycles: Alternate fresh independent code reviews with targeted fixes and validation, capped at three repair rounds plus one verification pass by default. - Model-diverse independent reviewers: Route each review to a different model family than the candidate producer, with context isolation so reviewers never see prior round conclusions. - Scoped authorization modes: Choose Local, Publish, or Respond mode to control whether the Skill edits locally, commits and pushes, or replies to review threads; merges and posted reviews always need separate explicit authorization. - Use Case: A reviewer leaves blocking feedback on your PR. Ask the agent to harden the PR, and it fixes credible findings, runs the repository's tests, re-reviews the exact new candidate with a different model, and reports convergence or a precise blocker. ## Quick Start Use harden-pr to review and fix the open pull request for this branch until it converges or hits the round budget.

Frequently Asked Questions about harden-pr

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

FAQPage Schema
How do I address all review feedback on a GitHub pull request automatically?▼

Run a hardening loop that classifies each finding as fix, defer, duplicate, or not applicable, repairs credible Blocking and Major issues, runs the repository's tests, and re-reviews the updated candidate. Each fix stays traceable to its finding in a round ledger.

How to get a second code review from a different AI model on my PR?▼

The Skill spawns a fresh reviewer subagent with an explicit model override and no inherited conversation history, preferring a model family different from the one that produced the candidate. If no alternative model is available, it discloses that the review was not cross-model.

Can it push commits or reply to review threads on my behalf?▼

Only with explicit authorization. Local mode edits the checked-out branch only, Publish mode adds ordinary commits and pushes, and Respond mode adds replies to verified threads. Merges, posted reviews, force-pushes, and summary comments each require separate approval.

When does the PR hardening loop stop?▼

It stops when a fresh review of the exact latest candidate finds no credible Blocking or Major issues and validation passes, or earlier when the round budget is exhausted, the same finding survives two repairs, or validation cannot pass without widening scope.

Will hardening a PR expand its scope or add new features?▼

No. Hardening only fixes defects within the PR's declared intent. Findings that require new capabilities, out-of-scope cases, or reinterpreted intent are recorded as deferred-out-of-scope rather than implemented, regardless of their stated severity.