How does Guardian learn from my feedback?

It tracks your acceptance and rejection rates for different types of suggestions. If you frequently reject style advice but accept security tips, Guardian will adjust its thresholds to focus on what you find most valuable.

Does it send my entire conversation to subagents?

No, Guardian uses a minimal context extraction pattern. It only passes the specific files and relevant Oracle patterns needed for a focused review, which saves tokens and improves response accuracy.

What triggers a Guardian intervention?

Guardian activates automatically when it detects high file churn, repeated errors, large code volume changes, or approaching context limits. You can also manually invoke it using commands like 'guardian review'.

Can Guardian modify my files directly?

No, Guardian and its subagents operate in a strictly read-only mode. They analyze code and provide suggestions, but all modifications require explicit user approval via the Guardian suggestion system.

Guardian Quality Monitor

Name: Guardian Quality Monitor
Author: Overlord-Z

byOverlord-Z

0•

Analytics & Monitoring

Monitors session health and automates code quality gates by spawning focused review agents when degradation or complexity thresholds are met.

Guardian acts as an intelligent safety net for your development sessions, working silently in the background to identify potential issues before they escalate. By monitoring metrics like code volume, repeated errors, and file churn, it automatically triggers read-only Haiku subagents to provide targeted code reviews and structured task planning. It integrates deeply with an Oracle knowledge base to validate suggestions against known patterns and dynamically adjusts its sensitivity based on user feedback, ensuring that interventions remain helpful, context-specific, and non-intrusive without ever modifying your codebase without approval.

Key Features

01Automated code quality reviews using lightweight, read-only Haiku agents

02Oracle-backed suggestion validation to prevent repetitive or conflicting advice

03Real-time session health monitoring including error rates and context bloat

04Minimal context extraction to ensure privacy and token efficiency

05Dynamic sensitivity adjustment based on user acceptance rates and history

060 GitHub stars

Use Cases

01Automating security and logic reviews for large blocks of newly written code

02Detecting and breaking out of repetitive error loops during debugging sessions

03Decomposing complex, multi-component architectural tasks into manageable plans

What are Skills?·How to Install

Install with 🐟 Skill.Fish

npx skillfish add overlord-z/claudeshack guardian

For use in Claude.ai and ChatGPT

Download Skill