Self-Reflection Agent Optimizer

소개

The Self-Reflection skill for contextd empowers Claude to perform recursive analysis on its own operational history. By mining memories and behavioral data, it identifies systemic patterns such as rationalized skips, overclaiming, and ignored instructions. It goes beyond simple error logging by performing causal chain analysis to find root causes in project documentation (CLAUDE.md) and provides pressure-tested remediations. This skill is essential for maintaining a healthy ReasoningBank and ensuring that agent behavior remains aligned with project requirements over long-term development cycles.

주요 기능

• Causal Chain Analysis for tracing symptoms to root-cause documentation gaps
• Pressure-tested remediation flow using subagents to verify documentation fixes
• 0 GitHub stars
• ReasoningBank health monitoring for memory quality and tag hygiene
• Behavioral Taxonomy Reporting to categorize and prioritize agent failures
• Predictive risk alerts for high-complexity or security-adjacent tasks

사용 사례

• Conducting periodic audits of agent behavior to prevent instruction drift
• Improving CLAUDE.md after a series of failures or poor project outcomes
• Predicting and mitigating risks before starting complex, multi-step code refactors