cost-aggregation

name	cost-aggregation
description	Consolidates provider-specific AI costs into unified metrics across Anthropic, OpenAI, Google Gemini, GitHub Copilot, and Ollama. Enables apples-to-apples cost comparison and savings analysis.
skill	cost-aggregation
version	1.0.0
task_id	COST-002
depends_on	["COST-001"]
author	senior-engineer
status	active

Cost Aggregation Skill (COST-002)

Consolidates provider-specific AI costs into unified metrics across 5 providers: Anthropic, OpenAI, Google Gemini, GitHub Copilot, and Ollama (local/zero-cost).

Purpose

Enable apples-to-apples cost comparison across AI providers for any given task, supporting model selection optimization (COST-003) and budget enforcement (COST-001).

API

from src.skills.cost_aggregation.scripts.cost_aggregator import CostAggregator

agg = CostAggregator()

# Compare cost across providers for a task
result = agg.aggregate_task_cost(
    task_type="code_review",
    input_tokens=5000,
    output_tokens=2000,
    model_variants={
        "anthropic": "claude-sonnet-4.6",
        "openai": "gpt-5.4",
        "google": "gemini-2.0",
        "copilot": "claude-sonnet-4.6",
        "ollama": "mistral:latest"
    }
)
# Returns per-provider cost, winner, and savings vs cheapest cloud

# Daily cost trend for a provider
trend = agg.cost_trend_for_provider("anthropic", "2026-05-01", "2026-05-28")
# Returns daily_spend list, total, avg_per_day

# Provider health check
health = agg.provider_health_check()
# Returns status per provider (healthy/degraded/unknown)

# Record actual usage for trend tracking
agg.record_usage("anthropic", "claude-sonnet-4.6", 5000, 2000)

Providers

Provider	Auth	Cost Model	Zero-Cost
anthropic	ANTHROPIC_API_KEY	per-token	No
openai	OPENAI_API_KEY	per-token	No
google	GOOGLE_API_KEY	per-token	No
copilot	GITHUB_TOKEN	per-use	No
ollama	(none)	local	Yes

Configuration

Provider pricing is configured in src/config/providers.yaml. Update that file when providers change their published rates.

Performance

• aggregate_task_cost(): <10ms (pure computation, no network calls)
• provider_health_check(): <1ms when cached (5-minute TTL)
• Cost accuracy: ±2% vs published rates

Files

src/skills/cost-aggregation/
├── SKILL.md                          # This file
├── scripts/
│   ├── __init__.py
│   ├── cost_aggregator.py            # Main CostAggregator class
│   └── providers/
│       ├── __init__.py
│       ├── base_provider.py          # Abstract base class
│       ├── anthropic_provider.py     # Anthropic claude-* models
│       ├── openai_provider.py        # OpenAI GPT-* models
│       ├── google_provider.py        # Google Gemini models
│       ├── copilot_provider.py       # GitHub Copilot (per-use)
│       └── ollama_provider.py        # Ollama local (zero-cost)
└── tests/
    ├── __init__.py
    └── test_cost_aggregator.py       # 80+ test cases

Dependencies

• COST-001 (cost_budgeter): Uses same rate structure pattern
• pyyaml: providers.yaml loading
• pytest: Test suite

Self-Improvement

This skill participates in the framework's continuous improvement cycle (see skill-improvement-feedback).

When you use cost-aggregation during a task, include a skill_feedback entry in your HANDBACK to help improve it over time:

skill_feedback:
  - skill_name: cost-aggregation
    effectiveness_score: 0.85        # required: 0.0–1.0
    clarity_score: 0.90              # optional
    coverage_gaps:
      - "Specific scenario the skill did not address"
    improvement_suggestions:
      - "Concrete change that would have helped"
    usage_context: "One sentence on how you used this skill"

Positive feedback is as valuable as critical feedback. Three or more feedback items for this skill automatically trigger an improvement task.