Context

Implement the promotion and demotion system defined in spec 8.4 that adjusts agent capabilities based on sustained performance.

Promotion criteria:

• Sustained high quality scores over time
• Successfully handling increasing task complexity
• Positive peer feedback and collaboration scores

Demotion criteria:

• Repeated task failures
• Quality score drops below threshold
• Cost inefficiency (high cost relative to output quality)

Acceptance Criteria

• Promotion criteria evaluation engine (configurable thresholds)
• Demotion criteria evaluation engine (configurable thresholds)
• Promotion unlocks higher tool access (integration with progressive trust system)
• Model upgrades on promotion (e.g., agent moved to more capable LLM)
• Model downgrades on demotion (e.g., agent moved to cheaper LLM)
• Approval workflow for promotions (configurable: auto or human-approved)
• Notification to agent and team on promotion/demotion
• Cooldown period after demotion (prevent rapid flip-flopping)
• Promotion/demotion history tracking
• Unit tests for criteria evaluation and state transitions

Dependencies

• Implement CFO agent logic (spending monitoring, model downgrade suggestions) #46 — Performance tracking must be implemented
• Implement progressive trust system with pluggable TrustStrategy protocol (DESIGN_SPEC §11.3) #43 — Progressive trust system for tool access integration

Design Spec Reference

• Spec 8.4 — Promotion and demotion system

Design Decisions Finalized

• D2 — Quality Scoring: Pluggable QualityScoringStrategy protocol (gates quality_score_min thresholds).
• D3 — Collaboration Scoring: Pluggable CollaborationScoringStrategy protocol (gates collaboration_score thresholds).
• D13 — Promotion Criteria: Pluggable PromotionCriteriaStrategy protocol. Initial: configurable threshold gates (ThresholdEvaluator with min_criteria_met N of M). Default: junior→mid = 2/3, mid→senior = all.
• D14 — Promotion Approval: Pluggable PromotionApprovalStrategy protocol. Initial: senior+ requires human approval. Junior→mid auto-promotes. Demotions: auto-apply for cost-saving, human for authority-reducing.
• D15 — Model Mapping: Pluggable ModelMappingStrategy protocol. Initial: model follows seniority (default ON, opt-out via hr.promotions.model_follows_seniority). Changes at task boundaries only.

Common pattern: All strategies use pluggable protocol interfaces with one initial implementation. Alternative strategies are documented in DESIGN_SPEC.md for future.