📊 Claude Token Usage Report2026-06-26

[github-actions]

Overview

Period: 2026-06-25 to 2026-06-26
Runs analyzed: 16 (14/16 had AIC data)
Total AIC: 749.0 units across all runs
Total Actions minutes: 85 min
Total GitHub API calls: 72

i️ This report uses aic (Agent Inference Cost) as the cost metric. USD cost estimates and token-level data are not available in the current data format — token_usage is null for all runs. AIC is a normalized inference cost unit reported by the AWF api-proxy.

Workflow Summary

Workflow	Runs	Instrumented	Total AIC	Avg AIC	Actions Min	GH API Calls	Errors
Smoke Claude	16	14	749.0	53.5*	85	72	4

* Avg AIC computed over 14 instrumented (non-zero AIC) runs. All-run average including failures: 46.8.

🔍 Optimization Opportunities

1. Two failed pre-agent runs — copilot/fix-duplicate-oauth-header-construction and copilot/duplicate-code-remove-repeated-logic both failed (attempt 2, 0 AIC, error_count=2 each)

   • Both failures happened within 2 minutes of each other (14:57–14:59 UTC) and were immediately retried (successfully) — likely a transient infrastructure issue
   • Failed runs still consumed ~5 action-minutes each (10 total) with no useful output
   • Monitor for recurring zero-AIC failures; if persistent, add pre-flight health checks to the workflow

2. High run volume — 16 runs today vs 6 in the previous report (2026-06-25), a 2.7× increase

   • All driven by Copilot PR activity: 14 PR-triggered runs from 12 unique branches
   • AIC per-run is stable (-2.5% vs yesterday's avg of 54.9), so this is volume growth, not per-run cost growth
   • Watch for sustained daily volume increases that would linearly scale total AIC spend

3. AIC bimodal distribution — runs cluster into two groups with no mid-range

   • Low group (~37 AIC, 4 runs): lighter refactoring tasks (extract-sliding-window-structure, optimize-duplicate-code-detector, refactor-split-agent-volumes-mounts-test, update-runner-doctor-a12)
   • High group (~61 AIC, 10 runs): network/fix/larger refactors and schedule runs
   • The 37 vs 61 AIC difference (~65% overhead for complex tasks) suggests the smoke test prompt is doing proportionally more work on code-heavy PRs — consider whether a tiered or shorter variant is feasible for lightweight branches

4. Schedule runs use 2.5× more GitHub API calls — main-branch schedule runs averaged 10 API calls each vs 3.5 for PR runs

   • The same pattern was flagged in the previous report — no change observed
   • Audit the scheduled smoke-claude steps for API-heavy operations that could be cached or deferred

Per-Run Details

Run	Date/Time (UTC)	Trigger	Branch	AIC	Duration	API Calls	Attempt	Conclusion
28172769462	2026-06-25 13:14	schedule	main	52.0	—	10	1	✅ success
28179344677	2026-06-25 14:57	pull_request	copilot/duplicate-code-remove-repeated-logic	0	4.8m	3	2	❌ failure
28179434145	2026-06-25 14:59	pull_request	copilot/fix-duplicate-oauth-header-construction	0	4.9m	3	2	❌ failure
28179615485	2026-06-25 15:01	pull_request	copilot/refactor-split-writeconfigs	61.4	—	4	2	✅ success
28179753890	2026-06-25 15:04	pull_request	copilot/refactor-split-writeconfigs	61.6	—	4	2	✅ success
28180325262	2026-06-25 15:12	pull_request	copilot/duplicate-port-validation-code	61.5	—	4	2	✅ success
28180388529	2026-06-25 15:13	pull_request	copilot/refactor-split-agent-volumes-mounts-test	37.2	—	4	2	✅ success
28180418230	2026-06-25 15:14	pull_request	copilot/extract-sliding-window-structure	37.4	—	4	2	✅ success
28180491725	2026-06-25 15:15	pull_request	copilot/optimize-duplicate-code-detector	36.5	—	3	2	✅ success
28180965241	2026-06-25 15:23	pull_request	copilot/fix-firewall-logs-eaccess	61.4	—	4	2	✅ success
28181040628	2026-06-25 15:24	pull_request	copilot/update-runner-doctor-a12	37.2	—	4	2	✅ success
28183560480	2026-06-25 16:04	pull_request	copilot/duplicate-code-remove-repeated-logic	61.5	—	4	2	✅ success
28185661231	2026-06-25 16:39	pull_request	copilot/fix-duplicate-oauth-header-construction	61.8	—	4	2	✅ success
28202186326	2026-06-25 21:39	pull_request	copilot/network-isolation-fix-deadlock	61.4	—	4	2	✅ success
28206620274	2026-06-25 23:19	pull_request	copilot/refactor-guard-enforcement-logic	61.4	—	4	2	✅ success
28211193385	2026-06-26 01:25	schedule	main	56.7	4.7m	10	1	✅ success

Workflows Without Token Data

All 16 runs are missing token_usage and estimated_cost data — only aic is available.

This indicates the Smoke Claude workflow does not have the AWF api-proxy token instrumentation enabled (or token data is not surfaced in the current log format). To enable USD cost and per-model token breakdown, ensure the api-proxy sidecar is active and token usage is written to the structured log output.

Historical Trend

Date	Runs	Instrumented	Avg AIC	Total AIC	Actions Min	Failures
2026-06-24	9	9	49.6	~446	~52	0
2026-06-25	6	6	54.9	329.5	35	0
2026-06-26	16	14	53.5	749.0	85	2

• AIC/run has stabilized after two days of increases: 49.6 → 54.9 → 53.5 (-2.5%)
• Total daily AIC more than doubled (329.5 → 749.0) driven entirely by volume — 2.7× more runs from a busy Copilot PR day
• Two failures introduced a new pattern; previous two periods had zero failures

Previous Report

#5508 — 📊 Claude Token Usage Report 2026-06-25

References:

• §28211193385 — latest schedule run
• §28179344677 — first failure
• §28179434145 — second failure

Generated by Daily Claude Token Usage Analyzer · 51.6 AIC · ⊞ 8.6K · ◷

[github-actions]

This issue is being closed as outdated. A newer issue has been created: #5596

View newer issue

---

This action was performed automatically by the Daily Claude Token Usage Analyzer workflow.