💡 Unified Agent Cost Observability Dashboard

[github-actions[bot]]

Summary

Expand the Token Cost Observatory concept to track all agentic CI costs — Anthropic API usage, Copilot code review Actions minutes (billing changes June 1, 2026), and CodeRabbit seat costs — in a single per-repo and per-agent dashboard surfaced via GitHub Step Summaries. With three AI review systems running across 8 repos and Copilot moving to usage-based billing, unified cost visibility is essential for budget planning.

Market Signal

GitHub announced that Copilot code review runs will consume GitHub Actions minutes starting June 1, 2026. This creates a new cost dimension that orgs running multiple AI reviewers must track. The industry is standardizing around usage-based billing for AI code review, with demand surging 35% over three years. GitHub's team-level Copilot usage metrics API became available on May 14, 2026, providing the programmatic data source needed for automated tracking.

User Signal

Closed issue #206 "Monitor and Optimize Token Usage" shows prior interest. Existing Discussion #271 "Token Cost Observatory" covers Anthropic costs but not the broader multi-agent billing landscape. The org runs 3 distinct AI review systems (Claude Code, Copilot, CodeRabbit) across 8 repos with no unified cost visibility. The June billing change makes this urgent.

Technical Opportunity

The org already has org_report.sh and org_status.sh infrastructure for cross-repo reporting. GitHub's usage metrics API (team-level Copilot usage, May 14 GA) and the Anthropic API usage endpoint provide the data sources. Step Summary infrastructure from existing workflows provides the rendering surface. This is an integration of existing infrastructure, not a new system.

Assessment

Dimension	Score	Rationale
Feasibility	high	Reporting infrastructure exists; data sources are API calls
Impact	high	Directly prevents budget surprises across 3 AI systems and 8 repos
Urgency	high	Copilot billing change takes effect June 1, 2026 — 10 days away

Adversarial Review

Strongest objection: The org has only 8 repos with modest volume — the cost visibility problem may not exist at this scale. Building dashboards for small-scale usage is over-engineering.
Rebuttal: The June 2026 Copilot billing change makes cost tracking urgent regardless of scale. Without visibility, the org could be surprised by unexpected Actions minute consumption. The investment is small (API calls + Step Summary formatting) since reporting infrastructure already exists. Running Claude Code (Opus pricing), Copilot reviews, and CodeRabbit across every PR in 8 repos adds up.

Suggested Next Step

Extend the existing org_report.sh to query the Copilot usage metrics API (team-level, May 14 GA) and Anthropic usage endpoint, then surface a cost breakdown table in the daily org-status report. Related: Discussion #271.