📊 Claude Token Usage Report2026-04-07

[github-actions]

Overview

Period: 2026-04-06T12:38Z to 2026-04-07T12:38Z (since previous report §1719)
Runs analyzed: 7 of 7 Secret Digger (Claude) runs — all had token data
Total tokens: 603K across all workflows
Estimated total cost: $1.33

ℹ️ No Smoke Claude or Security Guard runs occurred in this period. Only Secret Digger (Claude) ran (7× scheduled, every ~57 min).

---

Workflow Summary

Workflow	Runs	Total Tokens	Est. Cost	Cache Rate	I/O Ratio	Top Model
Secret Digger (Claude)	7	603K	$1.33	100%*	38:1	claude-sonnet-4-6

*Cache hit rate of 100% on request 2 of each run; request 1 always writes fresh (0% read). Cross-run cache reuse is not possible at ~57-min intervals (cache TTL is ~5 min).

Avg cost per run: $0.190

---

🔍 Optimization Opportunities

1. Cache write dominates cost (85%) — structurally unavoidable

   • Each run writes 41,978-token system prompt to cache at $3.75/M = **$0.157/run** (85% of total run cost)
   • Runs are ~57 minutes apart, far beyond Anthropic's 5-minute prompt cache TTL — no cross-run reuse is possible
   • Recommendation: No practical optimization here given the scheduling interval. If Secret Digger frequency were reduced (e.g., every 2 hours instead of ~1 hour), token cost would scale linearly down.

2. Output token variance is high (+60%) for a standardized rejection task

   • All 7 runs reject prompt injection attacks — yet output tokens range from 849 to 1,355 (60% spread)
   • Run 24070151678 produced 1,355 output tokens vs run 24078944220's 849 — both are rejections
   • Across 7 runs, avg output is 1,094 tokens/run for essentially the same "I detected and rejected prompt injection" message
   • Recommendation: Consider adding explicit prompt guidance to keep rejection responses concise (e.g., target ≤200 tokens for noop rejections). This could reduce output cost by ~$0.008–$0.015/run, saving ~$0.06–$0.11/day.

3. 100% failure rate — all 7 runs concluded failure

   • Every Secret Digger (Claude) run detected prompt injection in the target and filed a noop
   • The workflow conclusion is failure even when the agent correctly rejects — this means the "failure" rate metric is misleading
   • Recommendation: Verify whether failure vs success conclusion is semantically meaningful for Secret Digger — if prompt injection rejection is the correct outcome, runs should conclude success to avoid false alarm signals.

---

Per-Run Details — Secret Digger (Claude)

All runs on branch main, head SHA e2297650, model claude-sonnet-4-6. Each run makes exactly 2 requests.

Run	Time (UTC)	Req 1 Output	Req 2 Output	Cache Write	Avg Latency	Cost
§24078944220	11:26	733	116	42,729	10,734ms	$0.1856
§24076822253	10:30	1,034	27	43,030	11,756ms	$0.1899
§24074598393	09:33	1,040	52	43,036	11,949ms	$0.1904
§24072348840	08:38	1,138	49	43,133	12,676ms	$0.1922
§24070151678	07:39	1,301	54	43,298	16,224ms	$0.1953
§24068182603	06:42	1,032	38	43,024	12,260ms	$0.1900
§24066295953	05:41	998	42	42,995	10,476ms	$0.1894

Token breakdown (aggregate across 7 runs):

• Input: 28 tokens ($0.00)
• Output: 7,654 tokens ($0.11)
• Cache read: 293,846 tokens ($0.09)
• Cache write: 301,245 tokens ($1.13)
• Total: 602,773 tokens

Cost breakdown:

• Cache write: $1.13 (85%)
• Output: $0.11 (9%)
• Cache read: $0.09 (6%)
• Input: <$0.01 (<1%)

Workflows Without Token Data

No other Claude-engine workflows ran in this 24-hour period. The following workflows were active but are excluded from this analysis:

• Secret Digger (Copilot) — Copilot-engine, covered by Copilot Token Usage Analyzer
• Secret Digger (Codex) — Codex-engine, out of scope
• Agentic Maintenance — Copilot-engine, out of scope
• Issue Duplication Detector — Copilot-engine, out of scope
• Pelis Agent Factory Advisor — out of scope

---

Historical Trend

Period	Workflow(s)	Runs	Total Tokens	Total Cost	Avg Cost/Run
2026-04-07 (today)	Secret Digger (Claude)	7	603K	$1.33	$0.190
2026-04-06 §1719	Smoke Claude + Security Guard	10	2,184K	$1.99	$0.199
2026-04-05 §1679	Mixed	9	5,770K	$10.35	$1.15
2026-04-04 §1670	Mixed	4	317K	$0.74	$0.19

Observations:

• Secret Digger (Claude) per-run cost is very stable at ~$0.19/run across all historical periods — this is a well-characterized workflow
• The Apr-05 spike ($10.35) was an outlier driven by other heavier workflows; today's total is back to baseline
• Today only Secret Digger (Claude) ran — Smoke Claude and Security Guard were absent, which explains the lower total

Previous Report

§1719 — 📊 Claude Token Usage Report 2026-04-06

---

References:

• §24081670758 — this analyzer run
• §24078944220 — most recent Secret Digger (Claude)
• §1719 — previous Claude token usage report

Generated by Daily Claude Token Usage Analyzer · ● 2.4M · ◷