ai-agents-meta-orchestratorlisted

# AI Agents Meta Orchestrator The single entry skill for *building agents that build* — the meta layer above any one agent app. It locates the task on the **lifecycle × concern** map (plan → compose → orchestrate → loop → audit → economize → evolve) and delegates to one of 14 specialist spokes. The cross-cutting decision every spoke turns on — **eval-first execution with a default-deny autonomy boundary** — lives in `ai-agents-meta-core`; read it before wiring an autonomous loop, granting tools, or routing by model cost. ## Cluster map (intent → spoke) **Foundation & discipline** - `agentic-engineering` — the core practice: eval-first execution, decomposition, cost-aware model routing. - `search-first` — research-before-coding; find existing tools/libs/patterns before writing custom agent code. **Plan & compose** - `blueprint` — turn a one-line objective into a multi-session construction plan; each step has a cold-start brief, dependency graph, and adversarial review gate. - `plan-orchestrate` — read a plan, decompose into steps, and emit ready-to-paste per-step agent chains for an orchestrator command. **Architect the runtime** - `agentic-os` — build a persistent multi-agent OS on Claude Code: kernel, specialist agents, slash commands, file-based memory, scheduled automation, no external DB. - `dynamic-workflow-mode` — design task-local harnesses, eval gates, and reusable skill extraction for adaptive agent runs. **Orchestrate & loop** - `team-agent-orchestration` — ru