Experienced LLMs
Self-education for AI coding agents. Learns from daily feedback and mistake corrections to make local agents more tailored to you over time.
Overview
Experienced LLMs adds self-education to AI coding agents (Antigravity, Cursor, Copilot, Claude Code). It gives AI a tool to learn from daily feedback and mistake corrections, compiling them into high-density operating directives without replacing AGENTS.md, Skills, or MCPs.
Key Principles
- Not AI Memory: Focuses on self-education and operational feedback loops, tailoring agents to your individual workflow over time.
- Privacy-First (Two Files Only): Strict local containment to
~/.experienced-llms/memory.dband~/.experienced-llms/EXPERIENCE.md. Zero telemetry, zero repo scanning, zero cloud phone-home. - LLM-Only Consolidation: Semantic deduplication, contradiction resolution, and superseding executed purely through LLM reasoning without brittle heuristic fallbacks.
- Dual Execution Pathways: Headless scheduled execution via CLI tool bridges (
--llm "<cmd>") or scheduled agent sessions using the--prepareand--applylifecycle. - AgenticSkills Compliance: Standardized on open agent skill specifications (
skills/experienced-llm/SKILL.md) for instant discovery. - Full-Stack Migration Framework: Rails and Omakub-style migration runner managing schema, agent skills, CLI scripts, and configuration updates.