substrate-selection

Route agentic coding tasks by applying a grounded selection matrix that separates orchestrator, delegated harness, and leaf executor lanes, ensuring runtime policy matches the correct coding backend without conflating provider differences with harness guarantees.

Coding workflows break during provider swaps because unverified runtime choices tangle orchestration layers with model differences. Distinguishing provider deltas from harness runtime guarantees and applying compensating controls prevents this breakage.

A model provider defines capability and token limits, while a runtime harness provides execution guarantees like bounded retries, timeouts, and observability. Separating these prevents mistakenly attributing harness limitations to the model.

Add compensating controls for weaker coding harnesses by implementing structured response packets, bounded retries and timeouts, observable progress signals, and stall escalation protocols directly within your skills or configuration.

You need explicit lane contracts for agent execution whenever you run bounded delegated tasks or manager-directed editing, ensuring each run records orchestrator, external harness, and leaf-agent lanes with concrete endpoint bindings for inspectability.

Yes, you can apply the substrate selection matrix to safely swap coding backends like aider or Copilot by cleanly separating the orchestrator substrate from the skill surface, preventing policies from becoming accidental runtime invocations.