Context

Implement the performance tracking system defined in spec 8.3 that collects and aggregates metrics for each agent. These metrics drive hiring/firing decisions, promotions, and progressive trust.

Metrics to track:

• tasks_completed / tasks_failed — Task completion counts
• avg_quality_score — Average quality rating across completed tasks
• avg_cost_per_task — Average LLM/tool cost per task
• avg_completion_time — Average time from assignment to completion
• collaboration_score — Effectiveness in multi-agent interactions

Acceptance Criteria

• Performance metrics model matching spec 8.3 fields
• Automatic metric collection triggered by task completion events
• Quality scoring from code reviews, output evaluation, and peer feedback
• Cost efficiency tracking (tokens used, tool invocations, total cost per task)
• Time tracking (assignment to completion duration)
• Collaboration score computation (message quality, helpfulness ratings)
• Queryable by agent, department, project, and time range
• Rolling averages and trend detection (improving/declining)
• Metrics persistence (survives restarts)
• Unit tests for metric collection and aggregation

Dependencies

• Implement HR engine (hiring, firing, onboarding, offboarding flows) #45 — HR engine must be implemented
• Initialize project structure with pyproject.toml and src layout #20 — Task management system for completion events

Design Spec Reference

• Spec 8.3 — Agent performance metrics

Design Decisions Finalized

• D2 — Quality Scoring: Pluggable QualityScoringStrategy — layered combination as initial implementation.
• D3 — Collaboration Scoring: Pluggable CollaborationScoringStrategy — automated behavioral telemetry as initial.
• D11 — Rolling Windows: Pluggable MetricsWindowStrategy protocol. Initial: multiple simultaneous windows (7d, 30d, 90d). Min 5 data points per window; below that, report "insufficient data."
• D12 — Trend Detection: Pluggable TrendDetectionStrategy protocol. Initial: Theil-Sen regression slope per window + configurable thresholds (improving/stable/declining). Min 5 data points.

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