AI Harness Engineering

Design a bounded, testable coding-agent loop with task contracts, repository context, independent controls, and reviewable delivery evidence.

Make repositories navigable for coding agents and reviewers with layered instructions, a compact map, runnable checks, and durable handoffs.

Constrain coding-agent execution with isolated workspaces, scoped identity, controlled egress, audit evidence, and safe recovery from policy denials.

Scale coding-agent delivery with task claims, isolated CI capacity, review-aware queues, deterministic gates, and measured autonomy.

Evaluate and observe coding agents with versioned cases, deterministic checks, privacy-aware OpenTelemetry events, redaction proof, budgets, and delivery evidence.