Can I preview changes before they are applied to my config?

Yes, you can use the --dry-run flag to view all optimization recommendations and token savings estimates without modifying any files.

Does the exclude pattern scanner work with different languages?

Yes, it automatically detects common generated files and build directories for various ecosystems including TypeScript, Go, Python, Dart, and more.

What are 'low-signal' rules in Technical.md?

These are generic platitudes that lack specific file paths, thresholds, or technical terms, often wasting space without providing actionable guidance to the AI reviewer.

What is the difference between 'lookup' and 'load' mode for the Decision Register?

'Load' mode sends the entire register to the LLM context, whereas 'lookup' mode queries specific entries as needed, which is much more token-efficient for large projects.

How does this skill save tokens during code reviews?

It identifies redundant or verbose rules in Technical.md and suggests excluding unnecessary files like build artifacts and source maps, significantly reducing the 'fixed context' sent to the LLM.

Candid Context Optimizer

Name: Candid Context Optimizer
Author: ron-myers

byron-myers

•

コラボレーションツール

Audits and optimizes the context loaded by the Candid code review plugin to reduce token waste and improve review precision.

Candid Context Optimizer is a specialized utility designed to streamline the Candid code review workflow by identifying and eliminating context bloat. It performs deep analysis on Technical.md guidelines, identifies redundant or low-signal rules, suggests essential file exclusion patterns, and tunes configuration settings for maximum efficiency. By providing a clear token budget breakdown and actionable recommendations, it ensures that LLM reviews remain high-quality while minimizing unnecessary token consumption and cost.

主な機能

01Optimization recommendations for Decision Register modes and pruning

02Smart scanning for missing exclude patterns like build artifacts and generated code

03Automated identification of verbose, duplicate, and low-signal rules in Technical.md

04Automated configuration tuning for branching strategies and project settings

05Detailed token budget estimation and visualization for context files

0615 GitHub stars

ユースケース

01Reducing context window noise to improve the relevance of AI code reviews

02Standardizing project configurations across multiple repositories using Candid

03Cleaning up legacy Technical.md files that have grown too large or repetitive

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add ron-myers/candid candid-optimize

For use in Claude.ai and ChatGPT

Download Skill