Hand off engineering work to Claude Code as an autonomous team or native dynamic workflow, with explicit Opus/Sonnet routing, foreground completion barriers, resumable sessions, repeatable JavaScript workflow phases, hooks, and worktree isolation. Use when another agent or CI needs Claude Code to implement a scoped task, run a codebase-wide audit or migration, execute research-build-verify-gap-close phases, launch work in the background, or install project subagents and `.claude/workflows`.
Own a software-engineering outcome end to end with a provider-routed agent fleet: research the real codebase, turn findings into implementation waves, edit safely, test and inspect the result, and persist through gap-closing cycles until the authorized goal is complete. Use when the user asks for a GPT engineer, autonomous engineer, complete feature or repository build, broad remediation, research plus implementation, multi-agent coding, explicit subagents, different models, or a durable /goal-style engineering run. Prefer confirmed Codex Sol, Terra, and Luna profiles or Claude model-specific profiles; preflight routing and never silently present same-model children as a model-diverse fleet.
Lead end-to-end engineering work with a capable main agent and a parallel fleet of model-pinned GPT-5.3-Codex-Spark subagents. Use when the user asks for a Spark fleet, ultra-fast parallel coding agents, rapid repository exploration, many bounded implementation shards, or low-latency independent verification while retaining architecture, integration, and final acceptance with the main agent.
Autonomously pursue a sustained engineering outcome through repeated research, planning, multi-agent implementation, integration, verification, and gap-closing cycles. Use when the user asks for a /goal-style run, says do not stop, finish the whole codebase, research and build autonomously, babysit an outcome, or wants the orchestrator to keep working across continuations until genuinely complete. Use native goal tracking when explicitly requested and available; otherwise maintain an equivalent goal ledger without inventing tool capabilities or broadening authority.
Turn an existing audit, finding list, issue set, review, failing-test report, or implementation plan into completed, verified code through a coordinated agent fleet. Use when the user asks to build from findings, implement every audit item, finish a known backlog, remediate review results, or continue from research without repeating the whole investigation. Validate each finding, preserve repository state, assign non-overlapping writers, integrate in dependency order, and track every item to an explicit disposition.
Coordinate hierarchical coding-agent fleets for repository-wide audits, implementation sprints, migrations, and complex work that benefits from parallel specialists. Use when a user asks for subagents, a fleet, parallel delegation, GPT-5.6 Sol, Terra, or Luna routing, broad codebase completion, or independent implementation and verification passes. Enforce bounded ownership, concurrency-aware waves, dirty-worktree safety, runtime-honest model handling, and evidence-based integration.