Question
What delegation behavior can this plugin actually rely on from OpenCode, and which parts of an Ultra adaptation would be host-enforced versus prompt-driven best effort?
Document from current OpenCode/plugin behavior and this repository:
- available task/subagent tools and how agents invoke them;
- model, reasoning, instruction, session, and account inheritance for child work;
- concurrency, cancellation, failure, nesting, and recursion controls;
- how plan, build, orchestrator, review, compact, and custom agents differ;
- how
collaborationProfileEnabled, orchestratorSubagentsEnabled, internal collaboration headers, and x-openai-subagent affect execution;
- whether the plugin can detect that delegation tooling is unavailable before a turn;
- gaps between Codex multi-agent V2 semantics and the OpenCode substrate.
The resolution should define the capabilities and constraints the product contract may safely promise, without deciding that contract.
Question
What delegation behavior can this plugin actually rely on from OpenCode, and which parts of an Ultra adaptation would be host-enforced versus prompt-driven best effort?
Document from current OpenCode/plugin behavior and this repository:
collaborationProfileEnabled,orchestratorSubagentsEnabled, internal collaboration headers, andx-openai-subagentaffect execution;The resolution should define the capabilities and constraints the product contract may safely promise, without deciding that contract.