📊 Claude Token Usage Report2026-04-15

[github-actions]

Overview

Period: 2026-04-14 to 2026-04-15
Runs analyzed: 25 (24 had token data; 1 cancelled before any tokens were consumed)
Total tokens: 5,119K raw (1,679K effective after Anthropic cache pricing)
Estimated total cost: $5.99
Total Actions minutes: 108 min
Total turns: 139
Total errors: 9 across 25 runs

Two Claude-engine workflows ran: Security Guard (12 runs) and Smoke Claude (13 runs). Both use claude-sonnet-4-6 with very high prompt cache efficiency (≥99.99%). Security Guard had the higher per-run cost and the most reliability issues (3 cancelled runs, 5 total errors).

---

Workflow Summary

Workflow	Runs	Total Tokens	Avg Tokens/Run	Total Cost	Avg Cost/Run	Total Turns
Security Guard	12	2,121K	177K	$3.23	$0.27	53
Smoke Claude	13	2,998K	231K	$2.76	$0.21	86
Total	25	5,119K	205K	$5.99	$0.24	139

---

🔍 Optimization Opportunities

1. Security Guard — 3 cancelled runs + 5 total errors (41.7% of runs with errors or cancellations)

   • Run #2172 cancelled with 3 errors and zero token usage
   • Runs #2163 and #2161 also cancelled
   • Error rate exceeds the 30% threshold — investigate reliability root cause

2. Security Guard — Lower cache read:write ratio (2.8:1) vs Smoke Claude (5.0:1)

   • Security Guard: 1,521K cache reads / 553K cache writes
   • Smoke Claude: 2,492K cache reads / 495K cache writes
   • Security Guard cache is being written more per reuse cycle — consider whether context windows are being reused across turns optimally

3. Smoke Claude — 2 failures out of 13 runs (15% failure rate)

   • Runs #2484 and #2497 both failed with 1 error each
   • Below the 30% threshold, but worth monitoring for regression

---

Per-Workflow Details

Security Guard

• Runs: 12 (#2157 #2159 #2160 #2161 #2162 #2163 #2164 #2166 #2171 #2172 #2173 #2174)
• Total tokens: 2,121K raw (avg 177K/run)
• Estimated cost: $3.23 (avg $0.27/run)
• Turns: 53 total (avg 4.4/run)
• Model: claude-sonnet-4-6 (Anthropic)
• Cache analysis: 1,521K cache reads / 553K cache writes → 2.8:1 read:write ratio; cache efficiency ≥99.99% on instrumented runs
• Error rate: 5 errors across 5 runs (3 cancelled, 2 failed) — ⚠️ >30% threshold exceeded

Run #	Conclusion	Tokens	Cost	Turns	Errors
#2157	✅ success	149K	$0.28	4	0
#2159	✅ success	248K	$0.44	6	0
#2160	✅ success	245K	$0.41	6	0
#2161	❌ cancelled	137K	$0.24	4	0
#2162	✅ success	231K	$0.37	4	0
#2163	❌ cancelled	130K	$0.14	4	2
#2164	✅ success	130K	$0.14	4	0
#2166	✅ success	218K	$0.29	5	0
#2171	✅ success	200K	$0.30	4	0
#2172	❌ cancelled	0	$0.00	0	3
#2173	✅ success	142K	$0.27	4	0
#2174	✅ success	291K	$0.35	8	0

---

Smoke Claude

• Runs: 13 (#2484 #2487 #2489 #2490 #2491 #2492 #2493 #2494 #2496 #2497 #2502 #2504 #2505)
• Total tokens: 2,998K raw (avg 231K/run)
• Estimated cost: $2.76 (avg $0.21/run)
• Turns: 86 total (avg 6.6/run)
• Model: claude-sonnet-4-6 (Anthropic)
• Cache analysis: 2,492K cache reads / 495K cache writes → 5.0:1 read:write ratio; cache efficiency 100% on all instrumented runs
• Error rate: 4 errors across 4 runs (1 cancelled, 2 failures) — 30.8%, slightly above threshold

Run #	Conclusion	Tokens	Cost	Turns	Errors
#2484	❌ failure	229K	$0.24	7	1
#2487	✅ success	230K	$0.24	6	0
#2489	✅ success	181K	$0.22	6	0
#2490	✅ success	227K	$0.24	7	0
#2491	✅ success	227K	$0.24	6	0
#2492	✅ success	273K	$0.16	7	0
#2493	❌ cancelled	182K	$0.13	6	2
#2494	✅ success	227K	$0.14	6	0
#2496	✅ success	271K	$0.16	7	0
#2497	❌ failure	182K	$0.24	7	1
#2502	✅ success	227K	$0.24	7	0
#2504	✅ success	271K	$0.26	7	0
#2505	✅ success	272K	$0.25	7	0

Runs Without Token Data

1 run had no token data (cancelled before any API calls):

• Security Guard run #2172 — cancelled, 3 errors, 0 tokens

---

Historical Trend

Period	Workflows	Runs	Total Cost	Avg Cost/Run	Total Tokens
Apr 14–15 (this report)	Security Guard, Smoke Claude	25	$5.99	$0.24	5,119K
Apr 5–13 (prev report)	Secret Digger, Security Guard, Smoke Claude	23	$6.81	$0.30	4,305K

Notable changes vs previous report:

• ✅ Avg cost/run dropped $0.30 → $0.24 (−20%) — likely because high-cost Secret Digger Claude workflow is absent from this period
• ⚠️ Security Guard gained 4 more runs; error/cancellation rate increased
• ➕ New: raw token volume increased (+19%) even without Secret Digger, driven by more Smoke Claude runs

Previous Report

§1967 — 📊 Claude Token Usage Report 2026-04-13

References:

• §24454997857 — current workflow run

Generated by Daily Claude Token Usage Analyzer · ● 540.2K · ◷