How does the dual-engine evaluation benefit the plan?

By running Claude and Codex in parallel, the skill cross-validates feasibility assessments to highlight areas of AI agreement or flag potential disagreements for human review.

Why should I use the planning skill instead of coding immediately?

It prevents technical debt and rework by validating your approach against real-world documentation and production code patterns before implementation begins.

Where are the planning artifacts stored?

All research, approaches, and evaluations are persisted in a local 'plans/' directory, allowing code review agents to verify the implementation against the original plan.

What sources does this skill use for research?

The skill concurrently queries Context7 for official documentation, Serper for web-based best practices, and GitHub for real-world production code examples.

Do I need Codex for this skill to work?

No, the skill automatically detects if Codex is available. If not, it proceeds with a comprehensive Claude-only evaluation while still providing full research data.

Evidence-Based Planning

Name: Evidence-Based Planning
Author: guyathomas

byguyathomas

0•

Productivity & Workflow

Conducts deep architectural research and cross-validated feasibility studies before implementing complex software features.

The Evidence-Based Planning skill transforms the development process by mandating rigorous research before a single line of code is written. It utilizes a multi-source approach—scanning library documentation via Context7, real-world implementations via Serper, and production codebases via GitHub—to formulate three distinct implementation strategies. Each approach is then cross-validated using a dual-engine evaluation involving Claude and Codex to identify potential risks and strengths, ensuring that final architectural decisions are backed by hard evidence and peer-reviewed by specialized AI models.

Key Features

01Multi-source research using Context7, Serper, and GitHub MCPs

02Dual-engine evaluation with Claude and Codex for feasibility validation

03Synthesized AI evaluation reports with confidence and risk mapping

04Automated formulation of exactly three distinct technical approaches

050 GitHub stars

06State persistence in plans/ directory for plan-to-review linkage

Use Cases

01Designing complex features requiring new architectural patterns or unfamiliar libraries

02Evaluating trade-offs between multiple implementation strategies for a specific requirement

03Ensuring production-grade standards by validating approaches against real-world GitHub codebases

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add guyathomas/workflows planning

For use in Claude.ai and ChatGPT

Download Skill