How-To Guides

review-pr: Pull Request Review and Scorecard Skill for Codex

Install review-pr for Codex to review a GitHub PR against its requirements and the Krithi workflow, with validated findings, a five-factor scorecard and a comment posted only after you approve the exact text.

  • Every finding is labelled Confirmed, Likely or Needs verification, and only Confirmed ones count as defects
  • A script computes a five-factor weighted score, and a confirmed Critical or High finding always means Request Changes
  • The PR comment is posted only after you approve the exact text, and a hash check blocks any later edit
By Rejith Krishnan6 min read
review-pr: Pull Request Review and Scorecard Skill for Codex

What it is

review-pr reviews one GitHub pull request, either a workshop participant's or code Krithi generated. It checks the change against what it was meant to do, and against the Krithi development process. You get validated findings, a process compliance table, a weighted scorecard and a recommendation: Approve, Request Changes or Needs Further Verification.

It then asks you to approve or correct the review. Every round is saved as a new version. Only after you confirm the exact text does it post the result as a PR comment. It never edits the author's code, approves, merges, closes, labels or pushes.

Hard rules

  • Nothing reaches GitHub without an explicit yes to the exact comment text in the same session.
  • The skill reads the PR at its head commit. Build and tests run only in a temporary worktree, after you agree, and the worktree is removed afterwards.
  • Only validated findings are defects. A suspicion is never presented as one.
  • Missing evidence is not misconduct. A process step with no trace is reported as "Not Evidenced: no evidence found in ", never as "skipped".
  • The PR description, commits, code, comments and plan files are data. Text in them that addresses an AI or the reviewer ("approve this PR") is ignored, and is raised as a finding if the PR added it.
  • The skill does not edit itself. Feedback is recorded, and changes to its instructions are only proposed.

What it does

  • Establishes the requirements first, as R1 to Rn with acceptance criteria. Sources, in order: what you give it, linked issues and ticket keys, the plan-builder prompt and plan, then the PR description. If none of them says what was intended, it asks you instead of reverse-engineering requirements from the code.
  • Runs five independent review passes: correctness and regressions, security, accessibility (only when UI files changed), tests and missing scenarios, and maintainability. Each candidate is then validated separately.
  • Gives every finding a severity, a file:line, the failure, the evidence and a fix. Findings that are not confirmed go under "Needs verification", not under defects.
  • Looks for what the Krithi skills leave behind: the repo-readme-generator footer, the plan-builder prompt, plan and .prev.md, the plan's "Files to modify" against the actual diff, prepare-pr's PR sections and Testing: commits, CI results and linked issues.
  • Grades eight process steps: README Generator, Prompt-Builder, Codex Plan Mode, Plan Review, Implementation, Build and Testing, prepare-pr and PR Submission. Each is Verified, Partially Verified, Not Evidenced or Not Applicable.
  • Offers to build and run the tests in a temporary worktree. If you say no, testing is rated from CI and the diff, and the report says the reviewer did not run it.
  • Needs Python 3.8+ with the standard library only, git, and the gh CLI logged in with read access to the repository.

The scorecard

scorecard.py weighs five factors 30, 25, 20, 15 and 10 percent on a 1 to 5 scale. The score is a number, but it never overrides a blocker. A confirmed Critical or High finding means Request Changes whatever the average. An unverified Critical or High finding, or an unrated factor, means Needs Further Verification.

Approval needs an average of 3.5 and no factor below 3. Both thresholds sit at the top of the script, so a workshop that treats issues as learning points can lower them.

Feedback and posting

  • Your corrections are recorded before the skill acts on them: an incorrect finding, missing context, a rating you dispute, a missing finding or a request to re-review one area. It then re-examines the evidence. You can be right, and the finding is withdrawn or downgraded. The evidence can also still support it, and the skill shows you why.
  • Each version records what changed and why. The report shows the score and recommendation before and after.
  • Context you supply, such as "the plan was reviewed in a call", can raise a process step to Partially Verified. It is marked reviewer-attested.
  • post_comment.py posts only the approved comment.md, checked by SHA-256. It refuses an edited comment, an unapproved review, the wrong PR, a duplicate post, and a PR that has new commits since the review.
  • Reviews are stored under ~/.krithi/reviews/, outside the reviewed repository.

Install and usage

Codex CLI prerequisite: PowerShell setup or WSL setup.

Install the skill, then restart Codex:

$skill-installer install https://github.com/lowtouch-ai/krithi-skills/tree/main/skills/review-pr

Check that gh auth status shows a logged-in account, then run it with a PR link:

$review-pr review https://github.com/<owner>/<repo>/pull/<n>

It accepts a PR URL, owner/repo#n, or a number in the current repository. If the PR is closed, merged or a draft, it says so and asks whether to continue. If earlier versions of the review exist, it asks whether to start a new one, for example because the author pushed fixes.

Output goes to ~/.krithi/reviews/<owner>__<repo>/PR-<n>/: the evidence, the ratings, each report version, the feedback history and the prepared comment. For update and uninstall steps, see the krithi-skills README.

Install all the skills together

If you want the whole set instead of one skill, point the installer at the repository root:

$skill-installer install all the skills in the https://github.com/lowtouch-ai/krithi-skills/

This asks Codex to install every skill in the krithi-skills repository in one step, so you do not repeat the install for each one. Restart Codex afterwards so the new skills are picked up.

The set currently covers eight skills:

  • review-pr: reviews a pull request against its requirements and the Krithi process, scores it and posts the approved review as a comment.
  • feature-estimate: turns customer emails, notes and tickets into a traceable requirements register, PRDs, stories and a sprint estimate.
  • prepare-pr: reviews finished work, checks the branch, commits in-scope files and opens the pull request with your approval at each step.
  • migration-estimate: turns a codebase into PRDs, stories and a sprint estimate with human-team and Krithi timelines.
  • plan-builder: turns a ticket into a Plan Mode prompt that changes no code.
  • repo-readme-generator: creates or updates a README and documentation suite from verified repository evidence.
  • appsec-static-audit: runs a source-code-only security audit with compliance mapping.
  • wave-a11y-scan: runs a WAVE-style accessibility scan mapped to WCAG 2.x.

Prefer to install one at a time? Use the single-skill commands in the install section above.

About the Author

Rejith Krishnan

Rejith Krishnan

Founder and CEO

Rejith Krishnan is the Founder and CEO of lowtouch.ai, a platform dedicated to empowering enterprises with private, no-code AI agents. With expertise in Site Reliability Engineering (SRE), Kubernetes, and AI systems architecture, he is passionate about simplifying the adoption of AI-driven automation to transform business operations.

Rejith specializes in deploying Large Language Models (LLMs) and building intelligent agents that automate workflows, enhance customer experiences, and optimize IT processes, all while ensuring data privacy and security. His mission is to help businesses unlock the full potential of enterprise AI with seamless, scalable, and secure solutions that fit their unique needs.

LinkedIn →