skills.rest
parasharcuraj
parasharcurajCOMMUNITY
Explore all skills by this author →
WebsiteGitHub

project-marlin

Evaluate AI-generated Python code changes on GitHub repositories through a six-step workflow.

Updated Aug 27, 2026
Search Tags:#workflow#code-review#ai-code-evaluation#pr-selection#prompt-preparation
One-click install
npx skills add https://github.com/parasharcuraj/project-marlin --skill project-marlin
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: project-marlin
Source: https://github.com/parasharcuraj/project-marlin/tree/main/.agents/skills/project-marlin
Command: npx skills add https://github.com/parasharcuraj/project-marlin --skill project-marlin

SYSTEM DOCUMENTATION & REQUIREMENTS

💡 This Skill includes references (resource) components.

What problem does it solve?

Project Marlin provides a structured, repeatable workflow for evaluating AI-generated Python code changes on real GitHub repositories. It guides you through assembling prompts, selecting meaningful PRs, running a three-turn evaluation, and rating and justifying outcomes, all while ensuring decisions are grounded in your own analysis and not external models.

Core Features & Use Cases

  • Six-step workflow: Assessment, PR selection, prompt preparation, CLI setup, PR creation, and submission, with strict sequential ordering.
  • Three-turn evaluation: Execute at least three deliberate turns (A vs B) and justify the better response.
  • PR-context-driven prompts: Create task-focused prompts that describe outcomes and behaviors, not implementation details.
  • Non-PR references: Avoid relying on existing PRs; work from a pre-PR baseline for reproducibility.
  • Production readiness mindset: Emphasize edge cases, acceptance criteria, and safe, auditable changes.

Quick Start

Follow Marlin's six-step workflow to evaluate an AI-generated PR on a real repository, starting with expert assessment and ending with submission.

Frequently Asked Questions about project-marlin

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

FAQPage Schema
How do I structure an AI code evaluation workflow for Python pull requests?▼

An AI code evaluation workflow for Python pull requests is structured as a six-step sequential process: expert assessment, PR selection, prompt preparation, CLI setup, PR creation, and submission, ensuring production-ready outcomes.

How does a three-turn interactive code evaluation work for AI-generated changes?▼

A three-turn interactive code evaluation executes at least three deliberate A versus B turns for AI-generated changes, requiring you to explicitly justify the better response and rate the final outcome.

Can I use external LLMs to assist with AI code review during the evaluation process?▼

You cannot use external LLMs during the AI code review evaluation process; the workflow enforces non-use of external models to ensure all decisions are grounded in your own analysis.

What is the best way to prepare prompts for evaluating AI-generated Python code?▼

The best way to prepare prompts for evaluating AI-generated Python code is to create task-focused prompts from a pre-PR baseline that describe desired outcomes and behaviors, avoiding reliance on existing PR implementation details.

Why should I avoid using existing PRs as context when selecting prompts for AI code evaluation?▼

You should avoid using existing PRs as context for AI code evaluation to maintain reproducibility; working from a pre-PR baseline ensures prompts focus on outcomes rather than implementation details.

Does an AI code evaluation workflow require explicit acceptance criteria for production readiness?▼

An AI code evaluation workflow requires explicit acceptance criteria for production readiness, emphasizing edge cases and safe, auditable changes to ensure production-ready outcomes.

Related Skills

Explore More Software Engineering Skills→
Nickalus12Nickalus12
2

perf-check

Identify performance hotspots and regressions in Loom-style codebases.

Community
Intermediate
flosrnflosrn

create-pr

Create and push pull requests with auto-generated titles and descriptions.

Community
Intermediate
whichguywhichguy

advisors

Prompt Hermes models via 'hermes chat -q' for analysis and synthesis.

Community
Advanced
julianobarbosajulianobarbosa
7

reviewing-code

Review code for quality, bugs, security issues, and best practices using Codex.

Community
Advanced