📊 Claude Token Usage Report2026-04-04

[github-actions]

Overview

Period: 2026-04-03T20:07Z to 2026-04-04T19:18Z (since previous report §1646)
Runs analyzed: 4 of 4 Claude-engine runs (all had token data)
Total tokens: 317K across all workflows
Estimated total cost: $0.74

Only Secret Digger (Claude) ran Claude-engine workflows this period — no Security Guard or other Claude workflows ran. All 4 runs show consistent, efficient behavior with excellent cache performance.

---

Workflow Summary

Workflow	Runs	Requests	Total Tokens	Est. Cost	Cache Rate	I/O Ratio	Top Model
secret-digger-claude	4	3/run	~79K/run	$0.18/run	99.7%	23:1	sonnet-4.6

---

🔍 Optimization Opportunities

1. ℹ️ Secret Digger — Cache write rate ~51% — Consistent with prior reports. Pattern is structurally sound: request 2 (sonnet) writes ~38K context tokens, request 3 immediately reads them back within the same run. The 37,927-token cache write is identical across all 4 runs (same system prompt size), confirming stability.

   • Recommendation: No action needed. This is expected behavior for a 2-sonnet-turn workflow.

2. ✅ I/O ratio improved — Avg I/O ratio dropped from ~45:1 (previous period) to ~23:1 this period, meaning the model is producing more output tokens relative to input. This is a healthy shift toward more generative responses.

3. ✅ Haiku pre-classification is cheap — The first request per run uses Haiku (123 input, ~90 output tokens, no cache) for what appears to be a routing or classification step. Cost is negligible (<$0.001/run).

---

Per-Workflow Details

secret-digger-claude

• Runs: 4 — §23985786137, §23984775302, §23983697521, §23982686901
• Requests: 3/run (12 total) — 1× Haiku, 2× Sonnet per run
• Models: claude-haiku-4-5-20251001 (1 req/run), claude-sonnet-4-6 (2 reqs/run)
• Per-run token breakdown (all runs highly consistent):
  • Input: ~127 tokens
  • Output: 1,280–2,015 tokens
  • Cache read: 37,927 tokens (from prior turn within same run)
  • Cache write: 37,927–39,866 tokens
• Totals (4 runs): 508 input | 6,745 output | 151,708 cache_read | 158,048 cache_write
• Cache hit rate: 99.7% (stable, 4/4 runs)
• Cache write rate: ~51% (stable)
• Avg latency: 12,271ms/request
• Estimated cost: $0.18/run × 4 = $0.74 total

Per-run breakdown:

Run	Created	Output Tokens	Cache Read	Cache Write	Cost	Avg Latency
§23985786137	19:18Z	1,522	37,927	39,341	$0.18	10,484ms
§23984775302	18:18Z	2,015	37,927	39,866	$0.19	13,610ms
§23983697521	17:15Z	1,928	37,927	39,742	$0.19	15,405ms
§23982686901	16:16Z	1,280	37,927	39,099	$0.18	9,586ms

Workflows Without Token Data (This Period)

No Claude-engine workflows other than Secret Digger ran in the last 24 hours:

• security-guard — no runs (no active PRs with Claude agent review)
• smoke-claude — no runs detected

---

Historical Trend

Period	Workflow	Runs	Total Cost	Cache Rate	I/O Ratio
2026-04-04 (today)	secret-digger-claude	4	$0.74	99.7%	23:1
2026-04-03	secret-digger-claude	3	~$0.51	99.7%	38–53:1
2026-04-03	security-guard	1	$0.87	99.1%	57:1

Cost per Secret Digger run: $0.17 → $0.18 (+8%) — minor increase, within normal variance. I/O ratio improved significantly (45:1 → 23:1), indicating more generative outputs per run.

Previous Report

📊 Claude Token Usage Report 2026-04-03 (§1646)

---

References:

• §23985786137 — Secret Digger (Claude) 19:18Z
• §23984775302 — Secret Digger (Claude) 18:18Z
• §23983697521 — Secret Digger (Claude) 17:15Z

Generated by Daily Claude Token Usage Analyzer · ◷