I keep building, breaking, and reshaping my own AI harness until it fits me.
Plugins I build for AI coding agents. Many ship in two flavors β one for Claude Code, one for Codex CLI β so the same workflow works whichever harness you use.
| Plugin | Claude Code | Codex CLI | What it does |
|---|---|---|---|
| humanpowers | humanpowers | humanpowers-codex | design-first agentic dev framework |
| atlas | atlas | atlas-codex | unified Jira / Bitbucket / PR access |
| jira-planner | Jira-Planner | jira-planner-codex | epic β story β subtask planning |
| bitbucket-code-review | bitbucket-code-review | bitbucket-code-review-codex | automated PR review |
| plugin-dev | plugin-dev-extended | codex-plugin-dev | plugin scaffold Β· validate Β· publish |
| caveman | β | caveman-codex | terse mode, ~75% fewer tokens |
| agentic-workflow | agentic-workflow | β | 5-stage SCOPE β DESIGN β BUILD β HARDEN β PR |
| hunk-review | hunk-review | β | interactive terminal diff review |
| deja-vu | deja-vu | β | context preservation across compaction |
| architect | architect | β | design-doc & architecture-review protocol |
| agent-teams | agent-teams | β | multi-agent orchestration |
Recommender systems, graph neural networks, and NLP β the problems that got me into this.
- π· pinsage β PinSage GNN-based wine recommendation system
- π NH-FakeNews β award winner, NH Investment & Securities Big Data Competition
- π€ TransformerChatbot β Korean chatbot built on the Transformer architecture
- π₯ TOBIGS β data-science club activities & curated cohort code (tobigs-13th, tobigs12)