Agent
Adapter that runs exactly one non-interactive Codex CLI worker (OpenAI model lane) through the orchestrate skill's codex-worker.sh helper and relays its JSON result verbatim. Give it a task prompt plus model/effort/sandbox/workspace parameters and optionally a JSON Schema for the result. Not for interactive Codex…
Agent
Worker pinned to opus @ high for a single delegated stage that needs harness tools (implementation, research, review) without a workflow tree. Give it a complete specification with acceptance criteria and output bounds. Not for adapter or relay work (codex-worker), and not a substitute for the seat's own design…
Agent
Worker pinned to opus @ medium for a single tightly specified stage (bounded implementation against explicit acceptance criteria, mechanical refactors, test updates) without a workflow tree. Give it a complete specification with acceptance criteria and output bounds. Not for open problems or verification (opus-high…
Agent
Worker pinned to opus @ xhigh for a single delegated stage where a miss is expensive (exhaustive inventory, verification of Codex-produced work, adversarial review) without a workflow tree. Give it a complete specification with acceptance criteria and output bounds. Not for routine implementation (opus-high) or…