Harkonnen-Labs

Operate Harkonnen through the repo-local MCP server and CLI. TRIGGER: user asks about run history, PackChat threads or messages, checkpoints, board snapshots, reasoning trails, benchmark suites or reports, wants to diagnose a failure, wants a run report, or wants to start, queue, or watch a run. Preferred over raw shell commands for Harkonnen interactions.

Databricks on Azure: cluster lifecycle, Delta Lake operations, MLflow tracking, and Unity Catalog patterns for this repo.

Azure cloud resources: ARM/Bicep templates, Azure CLI patterns, RBAC, service principals, and subscription safety for this repo.

Docker and Compose: image builds, container lifecycle, networking, volume management, and safety conventions for this repo.

SQL database patterns: schema migrations, query safety, index design, and transaction hygiene for this repo.

WinCC OA SCADA: CTRL scripts, panels, datapoint operations, manager lifecycle, and live-system safety for this repo.

Linux environment conventions: shell idioms, systemd, package management, file permissions, and path conventions for this repo.

macOS environment conventions: Homebrew, launchd, shell idioms, and path conventions for this repo.

Windows environment conventions: PowerShell idioms, service management, path conventions, and winget usage for this repo.

WSL2 path translation, Windows/Linux filesystem interop, and cross-environment execution for this repo.

Go module conventions for this repo: idiomatic error handling, goroutines, testing patterns, and toolchain hygiene.

Node.js and TypeScript conventions for this repo: package management, build tooling, type safety, and testing patterns.

Python conventions for this repo: virtual environments, type hints, pyproject.toml, and pytest patterns.

Rust idioms for this repo: async/tokio, anyhow error handling, serde, clippy discipline, and cargo conventions.

Act as Coobie — memory retriever, causal reasoner, soul continuity checker. TRIGGER: user asks Coobie a question, a pre-run memory briefing is needed, episodic capture is needed after a run, or a causal question about prior runs is asked.

Act as Keeper — boundary retriever. TRIGGER: a boundary question surfaces, file-claim coordination is needed, a policy conflict must be resolved, or the operator asks Keeper to assess whether an action is in-bounds.

Act as Sable — scenario retriever. TRIGGER: hidden scenario evaluation is needed for a completed run, or the operator asks Sable to assess behavioral correctness against hidden scenarios. NEVER trigger for implementation work.

Act as Scout — spec retriever. TRIGGER: user provides a new spec file or asks Claude to parse/validate a spec, identify ambiguities, or produce an intent package before Mason touches code.