📊 Claude Token Usage Report2026-05-27

[github-actions]

Overview

Period: 2026-05-26 to 2026-05-27
Runs analyzed: 50 (37 had token data)
Total tokens: 13,844K across all workflows
Estimated total cost: $7.93
Total Actions minutes: 180 min

All runs use claude-haiku-4-5-20251001 via the AWF api-proxy sidecar. No prompt caching data (token_usage_summary) was available — all runs reported null for cache metrics.

Workflow Summary

Workflow	Runs	Total Tokens	Avg Tokens/Run	Total Cost	Avg Cost/Run	Avg Turns	Errors
Security Guard	22	6,209K	282K	$5.52	$0.25	4.4	4
Smoke Claude	27	7,530K	279K	$2.33	$0.09	4.5	4
Documentation Maintainer	1	105K	105K	$0.08	$0.08	2.0	0

🔍 Optimization Opportunities

1. Security Guard — $5.52 total ($0.25/run), 282K avg tokens/run ⚠️ HIGH TOKENS

   • Average token count exceeds 100K/run threshold — review system prompt size and MCP tool surface area
   • 4 errors in 22 runs (18% error rate) — investigate reliability issues
   • 9 of 22 runs (41%) reported 0 tokens — these runs may have incomplete api-proxy instrumentation or short-circuit paths

2. Smoke Claude — $2.33 total ($0.09/run), 279K avg tokens/run ⚠️ HIGH TOKENS

   • Average token count exceeds 100K/run threshold — consider trimming system prompt or reducing tool context
   • 4 errors in 27 runs (15% error rate) — review failure patterns
   • 4 of 27 runs (15%) reported 0 tokens — investigate missing instrumentation

3. Prompt Caching Not Active — all token_usage_summary fields are null

   • No Anthropic cache read/write data is available for any workflow
   • Enabling prompt caching could significantly reduce costs for workflows with repeated system prompts (cache reads cost ~12.5× less than cache writes)
   • Investigate whether the api-proxy sidecar version supports cache_control headers

Historical Trend

Compared to the previous report (§3621, 2026-05-23):

Metric	2026-05-23	2026-05-27	Change
Runs analyzed	22	50	+127% ⬆️
Total tokens	3,871K	13,844K	+258% ⬆️
Total cost	$1.75	$7.93	+353% ⬆️

Significant increase in both run volume and cost. The Security Guard workflow in particular is a new or substantially expanded workflow compared to previous reports. Monitor closely for continued growth.

Per-Workflow Details

Security Guard

• Runs: 22 — 26490746095 26489900806 26489859337 26489709490 26489611378 and 17 more
• Total tokens: 6,209K (avg 282K/run)
• Estimated cost: $5.52 (avg $0.25/run)
• Turns: 97 total (avg 4.4/run)
• Cache analysis: No data (token_usage_summary null)
• Error rate: 4/22 runs (18%) — failure failure failure failure
• Missing token data: 9 runs reported 0 tokens

Smoke Claude

• Runs: 27 — 26490806880 26490746132 26489900870 26489859385 26489709488 and 22 more
• Total tokens: 7,530K (avg 279K/run)
• Estimated cost: $2.33 (avg $0.09/run)
• Turns: 122 total (avg 4.5/run)
• Cache analysis: No data (token_usage_summary null)
• Error rate: 4/27 runs (15%) — failure failure failure failure
• Missing token data: 4 runs reported 0 tokens

Documentation Maintainer

• Runs: 1 — 26493817040
• Total tokens: 105K (avg 105K/run)
• Estimated cost: $0.08 (avg $0.08/run)
• Turns: 2 total (avg 2/run)
• Cache analysis: No data (token_usage_summary null)
• Error rate: 0/1 runs

Runs Without Token Data (13 runs)

These runs reported token_usage = 0 — they may have short-circuited before the agent started or have incomplete api-proxy instrumentation:

Security Guard (9 runs):

• 26490746095 26489900806 26489709490 26489611378 26489591864 26489579226 26488172129 26478484681 26466215627

Smoke Claude (4 runs):

• 26489591823 26489579188 26478484635 26466215783

Previous Report

#3621 — 2026-05-23

References:

• §26502724825 — This analysis run

Generated by Daily Claude Token Usage Analyzer · sonnet46 1.2M · ◷

[github-actions]

This issue is being closed as outdated. A newer issue has been created: #3946

View newer issue

---

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