#4 Mission + Autonomy

[diptimoyghosh9593-pixel]

Milestone. The 9 canonical missions as configs + the 2-level autonomy model.

Scope (build against #2's frozen schema + #3's Employee Runtime):

• Define the 9 missions as MissionDefinition configs (NOT 9 workflows; one Mission Runtime executes all): stalled-opportunity investigation, CRM hygiene review, follow-up coordination (RevOps); blocked-work investigation, roadmap/process health review, documentation sync (ProductOps); decision capture, SOP freshness review, ownership/bottleneck investigation (OrgIntelligence). Each config: id, employee, triggers, inputs, skills, capabilities, risk_policy (see YAML in the approved plan), goal, SOP link, KPIs (e.g. opportunity_age), authority, review cadence.
• Implement the 2-LEVEL autonomy model (they are DIFFERENT classifications):
  1. Work-design decision (Process Review): Automate / Delegate / Eliminate — what should happen to this piece of work.
  2. Runtime risk policy (per action): LOW / MEDIUM / HIGH / CRITICAL — how much authority the AI has. Cascade: Process Review -> Automate/Delegate/Eliminate -> if automated/delegated: Action -> Risk Classification -> Policy Decision -> Execute/Recommend/Approve/Block.
• Risk policy layer in apps/ai/src/policy/ (reuse trust battery/alert gate): classify action risk tier, wire to existing hitl/ + SignalWorkflow("hitl-approval"). Add policy_tier + rollback metadata to the action record.
• Lifecycle signals: pause/resume/redirect/reprioritize/replan via Temporal signals; every transition emits the frozen MISSION_* timeline verbs (extend timeline.py).

Tests (TDD): mission config loading + validation; autonomy cascade decisions for each risk tier; signal handling transitions.

Do NOT touch: prd.md/README.md, migrations (frozen). Coordinate with #3 on mission/employee.py ownership.