| name | copilot-usage-analytics |
| description | Use when estimating per-session Copilot CLI cost, analyzing model-routing efficiency, tracking agent dispatch patterns, or documenting which GitHub Copilot usage APIs exist. Covers session cost estimation, model selection recommendations, and API landscape mapping. |
| compatibility | ["VS Code","Cursor","Windsurf","Claude Code"] |
| metadata | {"category":"Uncategorized","tags":["uncategorized"],"maturity":"beta","audience":["developers"]} |
| allowed-tools | ["bash","git","grep","find"] |
Copilot Usage Analytics Skill
Use this skill when the goal is to estimate Copilot CLI session costs from dispatch patterns, recommend model-routing strategies based on task complexity, or document the current state of GitHub Copilot usage APIs.
When to Use
- Estimating cost for a completed or in-progress Copilot CLI session
- Analyzing which sub-agent dispatches are driving the most token consumption
- Choosing the right model tier for a task to minimize cost without sacrificing quality
- Auditing agent workflows for routing inefficiencies
- Generating an ROI report (issues resolved versus estimated spend)
- Documenting which Copilot usage APIs are available versus missing
How to Invoke
Reference this skill by attaching skills/copilot-usage-analytics/SKILL.md to your agent context, or instruct the agent:
Use the copilot-usage-analytics skill. Apply the session cost estimate template and the model-routing recommendation template to the current session's dispatch log.
Workflow
- Collect dispatch data — enumerate all tool calls and sub-agent dispatches in the session, noting the model identifier used for each.
- Estimate token consumption — use message length and interaction count as a proxy for input/output tokens per dispatch.
- Apply cost rates — map each model to its published token rate to produce a per-dispatch and per-session cost estimate.
- Identify high-cost dispatches — rank dispatches by estimated cost and flag any where a lighter model would suffice.
- Generate routing recommendations — for each flagged dispatch, recommend an alternative model tier and explain the trade-off.
- Produce session summary — fill in the session cost estimate template with totals, per-model breakdown, and optimization actions.
Templates in This Skill
| Template | Purpose |
|---|
templates/session-cost-estimate-template.md | Per-session cost breakdown by dispatch, model, and estimated token usage |
templates/model-routing-recommendation-template.md | Structured recommendations for right-sizing model selection per task type |
templates/api-landscape.md | Reference map of GitHub Copilot usage APIs — what exists, what is missing, and current workarounds |
API Landscape Summary
Three data sources were investigated for programmatic Copilot CLI usage data:
| Source | Endpoint | Status | Notes |
|---|
| GitHub REST — Copilot Usage | GET /orgs/{org}/copilot/usage | ⚠️ Partial | Returns IDE completion metrics only; CLI/agent token data not exposed |
| GitHub REST — Copilot Billing | GET /orgs/{org}/copilot/billing | ⚠️ Partial | Returns seat counts only; no per-model or per-session cost data |
| Power BI Copilot Usage dataset | Dataset 5c6c70ac-* | ❌ Auth error | Requires separate AAD scope (AADSTS9010010); not available via standard MCP token |
Until GitHub exposes per-session usage data (expected via Copilot Metrics API expansion), session cost must be self-tracked using this skill's estimation workflow.
Model Routing Guidance
| Task Complexity | Recommended Model | Rationale |
|---|
| Simple lookup, label assignment, short summarization | GPT-4o-mini / Claude Haiku | Low reasoning demand; high token efficiency |
| Code generation, refactoring, structured output | GPT-4o / Claude Sonnet | Balanced quality-to-cost ratio |
| Architecture design, multi-file reasoning, threat modeling | GPT-4o / Claude Sonnet 3.7+ | High accuracy required; cost justified by complexity |
| Creative or exploratory research | Claude Opus | Reserve for tasks where quality difference is measurable |
Guardrails
- Token estimates are proxies, not exact counts — present them with an explicit uncertainty range.
- Do not use model cost rates that are more than 30 days old; rates change and stale data misleads decisions.
- ROI reports must include the number of issues resolved, not just cost; cost alone is not an actionable metric.
- Do not make routing recommendations that sacrifice correctness for cost on security or compliance tasks.
Agent Pairing
This skill is designed to be used alongside:
- agentops agent (
agents/agentops.agent.md) — monitors agent lifecycle health and can incorporate cost signals into rollback and routing decisions.
- performance-analyst agent (
agents/performance-analyst.agent.md) — pairs token-cost data with latency and throughput metrics for a full efficiency picture.
- sprint-planner agent (
agents/sprint-planner.agent.md) — uses ROI estimates to prioritize automation investments.
