AI and platform engineering — open-source tools, agent infrastructure, and published research.
Blog · About · Report an issue
Practical AI-assisted software development tooling, built in the open. We share what we learn through reproducible experiments and writing at clouatre.ca.
- aptu — AI SDLC review harness for GitHub: structured context assembly for issue triage and PR review. Rust CLI, GitHub App, and GitHub Action.
- aptu-coder — On-device code-intelligence MCP server. Parses codebases into structural graphs so agents reason instead of re-deriving context.
- decisions-judge-mcp — Typed decisions for AI agents as an MCP tool: yes/no probabilities, choices, and scores on JSON state in one fast request.
Reproducible studies on LLM agents and pipelines, published with DOIs on Zenodo:
- llm-agent-experiments — open-weight models as drop-in replacements in multi-agent pipelines
- param-description-experiments — MCP parameter-description ablations
- prompt-repetition-experiments — instruction repetition in LLM-agent pipelines (null results, reported)
- rag-reranking-benchmarks — latency cost of cross-encoder reranking
- reversibility-benchmark — classifying irreversible agentic actions
- setup-goose-action · setup-kiro-action — GitHub Actions for AI agent CLIs in CI
- math-mcp-learning-server — educational MCP server for math and statistics
- homebrew-tap — Homebrew formulas for our tools
- Open an issue or discussion in the relevant repository
- Read and subscribe at clouatre.ca
