Agent tooling · Verification harnesses · Control planes · Quality infrastructure
AIを活用した研究・開発のための実行基盤、検証ハーネス、品質ツールを設計・実装しています。
QA / SDETを土台に、曖昧な要求を実行可能なワークフローへ落とし込み、コード分析、テスト、自動化、結果整理までを接続することが主な関心領域です。公開リポジトリでは、実際の開発で再利用できる単位へ分割したツールと運用資産を公開しています。
AI Conductor は、複数のAI、ツール、工程を目的に合わせて編成し、継続的な成果へつなげる立場を表す、自分なりの呼称です。
I build practical infrastructure for AI-assisted R&D: agent tooling, verification harnesses, control planes, and quality workflows.
以下は独立したデモの寄せ集めではなく、調査・要求整理から検証と品質判断までを支えるツール群です。
RanD → code-to-gate → HATE → manual-bb → QEG
Research & Requirements → Code Analysis → Automated Testing → Manual Acceptance → Quality Analysis
- domain-lakda-runner supplies runtime exploration and reproducible execution results.
- workflow-cookbook supplies reusable workflows, acceptance assets, and CI practices across the stack.
- shipyard-cp coordinates agent-assisted work across planning, development, acceptance, integration, and publishing.
| Layer | Repository | Responsibility |
|---|---|---|
| Research & Requirements | RanD | Research, hypothesis formation, requirement discovery, and acceptance framing. |
| Workflow Assets | workflow-cookbook | Task Seeds, acceptance workflows, reusable CI, and development practices. |
| Agent Operations | shipyard-cp | Coordinates agent-assisted work across planning, development, acceptance, integration, and publishing. |
| Code Analysis | code-to-gate | Converts source changes, static signals, architecture checks, and repository risk into reviewable results. |
| Runtime Exploration | domain-lakda-runner | Explores software behavior and produces reproducible runtime results. |
| Automated Testing | harness-auto-test-evidence | Normalizes automated test results for downstream use. |
| Manual Acceptance | manual-bb-test-harness | Supports risk-based manual black-box testing and acceptance review. |
| Quality Analysis | quality-evidence-graph | Connects requirements, risks, changes, tests, and review results for quality assessment. |
I am continuing to integrate and extend these tools through private R&D projects.
プロジェクト固有の構成、運用方式、進捗詳細は公開せず、単独でも再利用価値を持つ部分だけをOSSとして切り出しています。
- Build from concrete development and QA problems.
- Keep each repository independently understandable and usable.
- Include tests, examples, documentation, and CI with the implementation.
- Separate reusable public components from project-specific integration.
- Prefer working software and reproducible examples over broad claims.
Autonomous R&D Systems · Agent Tooling · Harness Engineering · Control Planes · Quality Infrastructure · Developer Workflows · Test Automation · LLM-native Development
- Languages: Python, TypeScript / JavaScript, SQL, Bash
- Interfaces & Data: JSON Schema, SQLite, REST APIs, CLI tools, structured outputs
- QA & Automation: pytest, Playwright, Airtest, Jest / Vitest, coverage, CodeQL, GitHub Actions
- Runtime & Infrastructure: Docker / OCI containers, devcontainers, async processing, local-first tooling
- LLM Systems: OpenAI-compatible APIs, local LLMs, llama.cpp / Ollama, routing and orchestration tooling