ai-pr-review_skill

This skill performs AI-powered PR reviews with inline comments, acceptance criteria verification, and optional fix-and-verify loops to streamline governance.

Python

9

GitHub Stars

1

Bundled Files

2 months ago

Catalog Refreshed

4 months ago

First Indexed

Readme & install

Copy the install command, review bundled files from the catalogue, and read any extended description pulled from the listing source.

Installation

Preview and clipboard use veilstrat where the catalogue uses aiagentskills.

npx veilstrat add skill vladm3105/aidoc-flow-framework --skill ai-pr-review

SKILL.md10.6 KB

Overview

This skill performs AI-powered pull request reviews with inline comments, severity classification, acceptance-criteria verification, and an optional fix-and-verify loop. It automates diff parsing, security and performance checks, and posts formal GitHub reviews with summary and conclusion metadata. The agent can apply PR labels and cross-post a review audit to linked issues.

How this skill works

The agent fetches PR diff and metadata, reads changed and contextual source files, and extracts linked issues from the PR body. It verifies acceptance criteria when an issue is linked, analyzes code for bugs, security, performance, and type-safety problems, then posts inline comments and a summary review via the GitHub API. Optionally, when fix-and-verify is enabled, it checks out the branch, applies limited fixes with co-authored commits, pushes changes, waits for CI, and re-reviews for up to three iterations.

When to use it

Automated reviews for new pull requests to catch regressions early
Verify that a PR meets linked issue acceptance criteria before merge
Enforce security, performance, and error-handling standards at PR granularity
Run a fix-and-verify loop for non-trivial REQUEST_CHANGES findings
Produce an auditable review record on linked issues and PR conversation

Best practices

Provide a clear PR description and link issues using "Closes/Fixes/Resolves #N" for acceptance verification
Enable fix-and-verify only when repository trust and CI policies allow automated commits
Keep changes small and focused to improve diff analysis accuracy
Exclude large binary or documentation-only files by using configured skip patterns
Treat AI reviews as advisory; require a human reviewer before merging

Example use cases

Run a full AI review on feature branches to catch security or logic bugs before human review
Verify that a bugfix PR satisfies acceptance criteria listed in the linked issue
Use low-only findings to suggest stylistic or API design improvements via COMMENT
Enable fix-and-verify on low- to medium-risk issues to accelerate iterative fixes
Create an audit trail by cross-posting review results to the linked issue for governance

FAQ

No. AI reviews are advisory. The agent posts a conclusion and labels but human approval is required to merge.

How many fix-and-verify iterations are allowed?

Fix-and-verify runs up to three iterations. If findings remain after the limit, the agent escalates to a human reviewer.