Phase 8: Goal-Driven Orchestration Architecture

Decision

Implemented strategic goal layer above tasks with 5-phase lifecycle (Discovery → Expert-Matched Decomposition → Expert Execution → Quality Integration → Goal Achievement). Goals have owner_agent_id (orchestrator), decompose into tasks via Agent Expertise Registry with [SKILL: x] markers, and are tracked via paperclip_list_goals on every orchestrator heartbeat. Cross-division goals use Brain handoffs. 9th Scaling Invariant (Goal-Driven Accountability) enforces ownership.

Rationale

User vision: strategic goals drive orchestrator behavior — orchestrators decompose goals into expert-matched tasks, quality-gate each delivery, integrate outputs, and confirm 100% achievement. Goals needed to sit above tasks in hierarchy with orchestrator ownership and full lifecycle management. Existing Paperclip Goals API (4 levels, 4 statuses) provided the data layer — needed MCP tools, agent instructions, templates, and Telegram integration to complete the stack.

Alternatives Rejected

Alternative 1: Goals as a task type (rejected — goals are fundamentally different from tasks, need separate lifecycle and ownership model). Alternative 2: Goal management only via Paperclip UI (rejected — Telegram bot needed for CTO workflow, orchestrators need MCP tools for autonomous goal monitoring).

Outcome

Pending