Can Guardian automatically edit my files?

No. Guardian sub-agents are strictly read-only. They analyze code and present suggestions for your approval, ensuring you retain full control over all modifications.

What triggers a Guardian intervention?

Guardian activates based on thresholds such as writing 50+ lines of code, repeating the same error three times, high file churn, or when context usage exceeds 70%.

What is the role of the Oracle integration?

The Oracle integration allows Guardian to validate suggestions against known project patterns and previous rejection reasons, preventing the AI from suggesting solutions you've already dismissed.

How does Guardian keep costs down?

Guardian uses the Haiku model for all background tasks and employs minimal context passing, only sending relevant code snippets rather than full conversation histories.

How does Guardian learn from my feedback?

It tracks your acceptance rates for different suggestion categories. If you frequently reject specific types of advice, Guardian increases its sensitivity thresholds or stops suggesting those patterns entirely.

Guardian Quality & Health Monitor

Name: Guardian Quality & Health Monitor
Author: Overlord-Z

byOverlord-Z

セキュリティとテスト

Monitors session health and codebase quality in real-time, automatically triggering focused sub-agents for code reviews and task planning.

概要

Guardian acts as an intelligent safety net for your Claude Code sessions by tracking file churn, error rates, and context usage to prevent logic degradation. When high-risk changes or repetitive errors are detected, it spawns lightweight, read-only Haiku agents to perform focused analysis and suggest improvements. By cross-referencing suggestions with an Oracle knowledge base and learning from your feedback, Guardian maintains high quality standards and provides actionable insights without compromising your authority over the codebase.

主な機能

Automatic triggers based on file churn, error repetition, and code volume
Read-only sub-agent reviews for secure, non-invasive code analysis
Real-time session health metrics to monitor context bloat and error rates
0 GitHub stars
Oracle knowledge base integration for pattern validation and consistency
Self-adjusting sensitivity that learns from user acceptance and rejection patterns

ユースケース

Detecting and course-correcting when a session begins to spiral due to repetitive errors
Performing automated security reviews on sensitive authentication or payment logic
Decomposing complex multi-component requirements into structured development plans

概要

主な機能

Automatic triggers based on file churn, error repetition, and code volume
Read-only sub-agent reviews for secure, non-invasive code analysis
Real-time session health metrics to monitor context bloat and error rates
0 GitHub stars
Oracle knowledge base integration for pattern validation and consistency
Self-adjusting sensitivity that learns from user acceptance and rejection patterns

ユースケース

Detecting and course-correcting when a session begins to spiral due to repetitive errors
Performing automated security reviews on sensitive authentication or payment logic
Decomposing complex multi-component requirements into structured development plans