📊 Claude Token Usage Report2026-05-17

[github-actions]

Overview

Period: 2026-05-16 to 2026-05-17
Runs analyzed: 48 (all 48 had token data)
Total tokens: 5,605K across all workflows
Estimated total cost: $2.84
Total Actions minutes: 116 min
Total turns: 178

Workflow Summary

Workflow	Runs	Total Tokens	Avg Tokens/Run	Total Cost	Avg Cost/Run	Avg Turns/Run	Errors
Smoke Claude	25	4,551K	182K	$1.65	$0.07	5.8	5
Security Guard	23	1,054K	46K	$1.19	$0.05	1.4	5

🔍 Optimization Opportunities

1. Smoke Claude — avg 182K tokens/run (above 100K threshold)

   • Token count is driven by large cache payloads (avg 148K cache-read tokens/run), which is generally good — but input + output tokens could be reduced by moving deterministic data-fetching to pre-agent steps
   • About 50% of turns per run appear to be data-gathering that could be pre-computed deterministically
   • Recommendation: Move data-fetching work to frontmatter pre-agent steps and write results to /tmp/gh-aw/agent/; this can cut inference turns by ~50%

2. Both workflows — error rate ~20%

   • Smoke Claude: 5/25 runs had errors (20%); Security Guard: 5/23 runs had errors (21.7%)
   • Both are below the 30% alert threshold but worth monitoring
   • Recommendation: Review error logs for systemic patterns; use awf logs stats to inspect blocked domains or tool failures

Per-Workflow Details

Smoke Claude

• Runs: 25 — latest
• Total tokens: 4,551K (avg 182K/run)
• Effective tokens: ~182K total → reflects heavy cache reuse
• Estimated cost: $1.65 (avg $0.07/run)
• Turns: 146 total (avg 5.8/run)
• Cache analysis: 3,696K cache-read tokens vs 816K cache-write tokens — excellent cache reuse ratio (~4.5×); cache prefix is being effectively reused across turns
• Models used: claude-haiku-4-5-20251001
• Error rate: 5/25 runs (20%) — below alert threshold
• Conclusions: 23 success, 2 cancelled

Security Guard

• Runs: 23 — latest
• Total tokens: 1,054K (avg 46K/run)
• Estimated cost: $1.19 (avg $0.05/run)
• Turns: 32 total (avg 1.4/run)
• Cache analysis: 833K cache-read tokens vs 214K cache-write tokens — strong cache reuse ratio (~3.9×)
• Error rate: 5/23 runs (21.7%) — below alert threshold
• Conclusions: 21 success, 2 cancelled

Workflows Without Token Data

All 48 runs had token data present. No workflows are missing instrumentation.

Historical Trend

This is the first Claude token usage report for this period. No previous report found for comparison.

References:

• §25977866073 — Smoke Claude (latest)
• §25976924896 — Security Guard (latest)
• §25986385476 — This analyzer run

Generated by Daily Claude Token Usage Analyzer · ● 2M · ◷

[github-actions]

This issue is being closed as outdated. A newer issue has been created: #3317

View newer issue

---

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