How does the verdict system work?

The schema uses four main verdicts: ACCEPT (valid issue), REJECT (incorrect finding), DEFER (valid but out of scope), and ACKNOWLEDGE (intentional design decision).

How does this help improve AI skills?

By analyzing 'REJECT' verdicts and their rationales, developers can pinpoint exactly where an AI skill is misunderstood or outdated, allowing for precise updates to the skill's logic.

What is the purpose of the Review Feedback Schema?

It provides a structured format for logging code review outcomes, which helps in identifying false positives and improving the accuracy of AI-driven code reviews over time.

Can I use this for custom linting rules?

Yes, the schema includes a 'rule_source' field specifically designed to track custom rules from different frameworks like Pydantic AI, FastAPI, or React.

Code Review Feedback Schema

Name: Code Review Feedback Schema
Author: existential-birds

byexistential-birds

•

Collaboration Tools

Standardizes the logging of code review outcomes to enable continuous improvement of AI-driven code analysis and verification.

About

The Review Feedback Schema provides a structured framework for capturing and analyzing code review results within Claude Code. By using a standardized CSV format to track findings, verdicts, and rationales, this skill enables developers to build a robust feedback loop. It allows for the identification of recurring false positives, tracking of skill accuracy over time, and the automated refinement of domain-specific review rules. It is an essential utility for teams looking to move beyond simple linting toward high-precision, data-driven code verification workflows.

Key Features

Structured CSV schema for consistent review outcome logging
Source-level tracking to identify specific rules causing false positives
Standardized verdict taxonomy (ACCEPT, REJECT, DEFER, ACKNOWLEDGE)
Guidance for writing high-quality rationales to improve AI training
Categorized issue taxonomy covering security, performance, and style
14 GitHub stars

Use Cases

Tracking the accuracy of automated code reviews to reduce developer friction
Auditing code quality trends across different repositories and frameworks
Creating a training dataset for refining custom project-specific coding rules

About

Key Features

Structured CSV schema for consistent review outcome logging
Source-level tracking to identify specific rules causing false positives
Standardized verdict taxonomy (ACCEPT, REJECT, DEFER, ACKNOWLEDGE)
Guidance for writing high-quality rationales to improve AI training
Categorized issue taxonomy covering security, performance, and style
14 GitHub stars

Use Cases

Tracking the accuracy of automated code reviews to reduce developer friction
Auditing code quality trends across different repositories and frameworks
Creating a training dataset for refining custom project-specific coding rules