Redesign Dispatch/Agent-Delegation Architecture (omoc swarm replacement)

[Delqhi]

Problem

The current OMOC swarm and dispatch delegation architecture is fragile and heavily relies on oh-my-opencode's hardcoded fallback chains. When an agent's primary model and limited fallbacks fail, the plugin defaults to xai/grok or anthropic/claude-haiku-4-5, leading to unwanted model usage and task failures. The user explicitly stated: "wir haben dispatch funktionen gebaut für omoc swarm usw aber das wurde gut gedacht und extrem, aber ultra schlecht umgesetzt! mach das von grundauf neu!"

Goal

Redesign the agent dispatch and delegation architecture from the ground up.

Steps

1. Exhaustive Config Overrides: Update oh-my-openagent.json with 4-5 fallback models per agent to guarantee we never hit the plugin's hardcoded layer. (Completed)
2. Skill Redesign: Refactor omoc-plan-swarm and check-plan-done to use robust, deterministic delegation.
3. Dispatch Logic: Implement explicit model/provider validation before delegation.
4. Live Verification: Ensure delegates always route to approved models (GPT-5.4, Antigravity Claude/Gemini, GLM-5.1, Qwen).

Subtasks

• Configure exhaustive oh-my-openagent.json
• Refactor call_omo_agent usage in swarm skills
• Test live dispatch delegation

[Delqhi]

Implementation Complete\n\n### Design Document\nSee docs/dispatch-architecture-v2.md for full architecture, rationale, and verification plan.\n\n### Changes Merged\n- Fork: oh-my-opencode-sin (empty AGENT_MODEL_REQUIREMENTS & CATEGORY_MODEL_REQUIREMENTS)\n- Config: oh-my-openagent.json with exhaustive fallbacks\n- opencode.json now loads local plugin path\n- Commit: 9a60a02\n\n### Next:\n- Run live verification on OCI VM (see docs for tests)\n- Update issue with test results