GPT Orchestration Auto

Use installed GPT Engineer as the shared contract. This wrapper adds durable progress, not another model policy. Preserve the selected gpt-6-astra, gpt-5.6-sol, gpt-5.6-terra, or gpt-5.6-luna parent; same-model children are explicit by default and mixed-model work is a deliberate policy. Native Claude execution remains provider-specific. No fleet is mandatory.

Record objective, scope, acceptance, authority, requirements, evidence, current stage and blockers. Use native goal tools only when explicitly requested and available, following their exact lifecycle and budget semantics. Otherwise use a compact private checkpoint; never claim automatic persistence or scheduled continuation without runtime support. Skills installation is not dependency resolution.

Carry forward authorization and steering within the active task. Routine decisions remain autonomous. Persistence does not authorize unlimited retries, spending, global setup, production writes, or new scope. Set stage/time/resource budgets before repeated work. On exhaustion, retain evidence and report remaining work; do not mark unfinished requirements complete or silently escalate the budget.