📊 Claude Token Usage Report2026-04-18

[github-actions]

Overview

Period: 2026-04-17 to 2026-04-18
Runs analyzed: 19 (all 19 had token data)
Total tokens: 4,302K across all workflows
Estimated total cost: $4.13
Total Actions minutes: 80 min
Total turns: 128

Workflow Summary

Workflow	Runs	Total Tokens	Avg Tokens	Total Cost	Avg Cost	Turns	Errors
Security Guard	9	2,758K	306K	$2.77	$0.31	67	2
Smoke Claude	10	1,544K	154K	$1.36	$0.14	61	0

🔍 Optimization Opportunities

1. Security Guard — $0.31/run avg, 306K tokens/run (above 100K threshold)

   • Token count is 3× above the 100K/run threshold; consider reducing system prompt size or limiting MCP tool surface
   • 2 of 9 runs failed (22% failure rate, near the 30% threshold) — the two failed runs also consumed the most tokens (388K, 352K), suggesting wasted cost on retries
   • The agentic assessment flags ~50% of turns as data-gathering that could move to deterministic pre-agent steps, potentially cutting cost nearly in half

2. Smoke Claude — cost spike on 2026-04-17 vs 2026-04-18

   • Runs on 2026-04-17 cost ~$0.20/run; runs on 2026-04-18 dropped to ~$0.07/run — a 3× improvement, likely due to improved prompt caching
   • Cache write tokens are nearly constant (~42K/run) but cache read tokens doubled from ~82K to ~125K in the same-day runs, confirming caching is working well

3. Security Guard cache writes vs reads

   • Cache write: 443K total, Cache read: 2,287K total → 5.2× read:write ratio — good reuse
   • However, the two failed runs show elevated cache writes (63K, 51K) with fewer cache reads proportionally, suggesting the failures occur early before cache warms up

Per-Workflow Details

Security Guard

• Runs: 9
  • 24612543343 2026-04-18 ✅ 278K tok $0.30 7 turns
  • 24610775084 2026-04-18 ✅ 321K tok $0.30 8 turns
  • 24610603509 2026-04-18 ✅ 357K tok $0.30 8 turns
  • 24593695636 2026-04-18 ✅ 419K tok $0.35 9 turns
  • 24593538042 2026-04-18 ✅ 279K tok $0.23 7 turns
  • 24593463088 2026-04-18 ✅ 231K tok $0.27 6 turns
  • 24591315615 2026-04-17 ❌ 353K tok $0.32 9 turns
  • 24590789067 2026-04-17 ❌ 389K tok $0.48 9 turns
  • 24586966581 2026-04-17 ✅ 131K tok $0.22 4 turns
• Total tokens: 2,758K (avg 306K/run)
• Estimated cost: $2.77 (avg $0.31/run)
• Turns: 67 total (avg 7.4/run)
• Model: claude-sonnet-4-6 (Anthropic)
• Cache analysis: write=443K, read=2,287K (5.2× reuse ratio — healthy)
• Error rate: 2/9 runs failed (22%)

Smoke Claude

• Runs: 10
  • 24612543356 2026-04-18 ✅ 168K tok $0.07 6 turns
  • 24610775074 2026-04-18 ✅ 126K tok $0.07 6 turns
  • 24610603484 2026-04-18 ✅ 169K tok $0.07 7 turns
  • 24605543422 2026-04-18 ✅ 166K tok $0.07 6 turns
  • 24593695609 2026-04-18 ✅ 127K tok $0.07 6 turns
  • 24593652310 2026-04-18 ✅ 122K tok $0.19 5 turns
  • 24593538065 2026-04-18 ✅ 208K tok $0.22 7 turns
  • 24591315622 2026-04-17 ✅ 166K tok $0.20 6 turns
  • 24590789047 2026-04-17 ✅ 166K tok $0.21 6 turns
  • 24586966583 2026-04-17 ✅ 125K tok $0.19 6 turns
• Total tokens: 1,544K (avg 154K/run)
• Estimated cost: $1.36 (avg $0.14/run)
• Turns: 61 total (avg 6.1/run)
• Model: claude-sonnet-4-6 (Anthropic)
• Cache analysis: write=421K, read=1,113K (2.6× reuse ratio — moderate; improving trend on 2026-04-18)
• Error rate: 0/10 runs failed (0%)

Workflows Without Token Data

All 19 runs had complete token data. No workflows are missing api-proxy instrumentation.

Historical Trend

This appears to be the first Claude token usage report (no previous claude-token-usage-report issues found).

References:

• §24612543343 — Security Guard (latest)
• §24612543356 — Smoke Claude (latest)
• §24590789067 — Security Guard (highest cost run)

Generated by Daily Claude Token Usage Analyzer · ● 225.3K · ◷