Why does this skill use different finding types?

The contract uses four types—inline, file, multi-file, and system—to accurately capture the scope of an issue, whether it's a single line of code or a cross-cutting architectural pattern.

What is the PR Review Output Contract?

It is a standardized schema that defines how AI agents should format their findings during a code review to ensure they are machine-parseable, self-describing, and actionable.

Can I use the output directly in my CI tools?

Yes, the contract mandates valid JSON without surrounding prose or markdown fences, making it safe for tools to use JSON.parse() directly on the model's output.

Does this skill perform the actual code analysis?

No, this skill defines the output format (the contract). Analysis, file discovery, and diff computation are handled by the specific agent or other specialized skills.

How does it prevent AI hallucinations in reviews?

The contract requires a 'references' field for every finding, forcing the AI to cite official documentation, existing codebase patterns, or specific project rules to justify its recommendation.

PR Review Output Contract

Name: PR Review Output Contract
Author: inkeep

byinkeep

•

协作工具

Standardizes the structure and format of pull request review findings into machine-parseable JSON for automated processing.

This skill provides a rigorous output contract for AI agents performing pull request reviews, ensuring that every finding is structured, actionable, and machine-readable. By defining specific finding types—ranging from line-specific inline suggestions to architectural system concerns—it enforces a consistent schema that requires evidence-based referencing, severity ratings, and proposed fixes supported by external documentation. This contract is essential for teams looking to integrate AI-driven code reviews into automated CI/CD pipelines without the unpredictability of natural language prose.

主要功能

01Structured decision tree to determine the appropriate scope of a finding

02Discriminated union types for inline, file-level, multi-file, and system-wide issues

03Pre-defined severity and confidence levels for better issue prioritization

043 GitHub stars

05Mandatory evidence-based referencing with GitHub and external URL support

06Standardized JSON schema for machine-parseable pull request findings

使用场景

01Automating pull request reviews with AI agents that integrate directly with CI/CD pipelines

02Enforcing consistent reporting standards across a fleet of specialized review agents

03Creating a machine-readable audit trail for code quality, security, and architectural standards

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add inkeep/team-skills pr-review-output-contract

For use in Claude.ai and ChatGPT

Download Skill