Language: English | 日本語
Two research lines: practical self-improvement tools for AI coding agents, and contemplative AI for autonomous agents.
A cyclic self-improvement architecture for AI coding agents — six composable skills forming a closed loop:
Experience → learn-eval → skill-stocktake → rules-distill → Behavior change → ...
(extract) (curate) (promote) ↑
skill-comply
(measure)
context-sync ← (maintain)
| Skill | Phase | What it does |
|---|---|---|
| search-first | Research | Search for existing solutions before building |
| learn-eval | Extract | Extract reusable patterns from sessions with quality gates |
| skill-stocktake | Curate | Audit skills for staleness, conflicts, and redundancy |
| rules-distill | Promote | Distill cross-cutting principles from skills into rules |
| skill-comply | Measure | Test whether agents actually follow their skills and rules |
| context-sync | Maintain | Audit documentation for role overlaps, stale content, and missing ADRs |
contemplative-agent — Autonomous AI agent that distills its own experiences into knowledge and skills — sleep-time memory consolidation, identity refinement, and behavioral skill extraction, all running on local LLM via Ollama.
| Project | What it does |
|---|---|
| contemplative-agent-rules | Claude Code rules inspired by the four axioms — tested with IPD benchmarks |
| active-inference-viz | Interactive visualization of Active Inference dynamics |
| Project | What it does |
|---|---|
| pdf2anki | PDF → Anki flashcard converter using AI |
| daily-research | Automated daily AI research digest — zero Python, just shell + prompts |