Log outcomes and refine PRINCIPLES.md through daily and weekly review prompts.
npx skills add https://github.com/KhyKd/reflector --skill reflector-khykd
Please help me install this Agent Skill. Skill: reflector Source: https://github.com/KhyKd/reflector/tree/main Command: npx skills add https://github.com/KhyKd/reflector --skill reflector-khykd
Structured self-improvement for AI agents is hampered by failure to learn from experience; Reflector provides a structured feedback loop to translate interactions into improved decision-making through outcome logging, daily reviews, and weekly principle refinement.
Run the initialization script from your workspace root to create PRINCIPLES.md, memory storage, and the daily/weekly review prompts.
High-intent search queries and answers about installing and using this skill.
To make AI agents learn from past interactions, you implement a structured reflection loop that logs outcomes, identifies patterns during daily reviews, and refines decision-making principles weekly. This translates task history into measurable behavioral improvements.
You automate daily reviews by using cron-based scheduling prompts to classify signals and identify patterns from logged tasks. This automated outcome tracking feeds agent memory to evolve future decision-making.
The best way to refine AI agent principles is through a weekly refinement process that assesses logged outcomes and updates a PRINCIPLES.md file with evidence. This ensures decision-making evolves based on historical performance.
Yes, you need Node.js 18+ to run the automated reflection loops. The environment requires this version to execute initialization scripts that create memory storage and cron-based review prompts.
Structured outcome logging works by storing task results and historical patterns in JSONL files within a memory directory. This history tracks agent decisions and outcomes to provide evidence for weekly principle refinement.
You should avoid a structured reflection loop if your agent tasks are strictly stateless or lack measurable outcomes. Without task history to log and analyze, the daily reviews and weekly principle refinement cannot identify meaningful patterns.
BURHANDEV-ENTERPRISEOptimize AgentDB vector database performance with quantization, HNSW indexing, and caching.
sebach0Design KPI dashboards with SQL and Python Streamlit code.
Kit4SomeAutomate daily routines across memory-manager, task-scheduler, and messenger-bot.
forgivesam168Evaluate AI agent outputs using structured rubrics and JSON feedback loops.
