Skip to content

📊 Claude Token Usage Report2026-04-05 #1679

@github-actions

Description

@github-actions

Overview

Period: 2026-04-04T20:03Z to 2026-04-05T19:58Z (since previous report §1670)
Runs analyzed: 9 Claude-engine runs (2 workflows had token data)
Total tokens: 5,770K across all workflows
Estimated total cost: $10.35 ⚠️ (primarily Daily Security Review via Copilot API — see note below)


Workflow Summary

Workflow Runs Total Tokens Est. Cost Cache Rate I/O Ratio Top Model
Daily Security Review 1 5,140K $8.91 † 48.8% 512:1 claude-sonnet-4.6
Secret Digger (Claude) 8 630K $1.44 99.7% 930:1 sonnet-4.6

Daily Security Review uses the Copilot API (provider: copilot). Cost estimate is based on Anthropic list pricing and may differ from actual billed amount, which goes through GitHub Copilot seat licensing.


🔍 Optimization Opportunities

  1. Daily Security Review — 48.8% cache hit rate, zero cache writes

    • All 33 requests use provider: copilot with no cache_write_tokens. Cache hits come from implicit Copilot session context reuse, not explicit Anthropic prompt caching.
    • The cache hit rate grows progressively within the run (earlier turns have low hit rates, later turns hit ~46K tokens), but the first call sees 0 cache hits at 40K input tokens.
    • Recommendation: If this workflow transitions to direct Anthropic API calls, enabling explicit cache breakpoints on the large system prompt could push the hit rate above 90%.
  2. Daily Security Review — High I/O ratio (512:1)

    • Average 79K input tokens vs 603 output tokens per request.
    • This is typical for deep analysis workflows, but the 33-request count with large contexts accumulates quickly ($8.91 in a single run).
    • Recommendation: Review whether the system prompt can be trimmed or whether some analysis steps can be consolidated.
  3. Several workflows missing instrumentation

    • Agentic Maintenance, Firewall Issue Dispatcher, and Documentation Maintainer all ran but produced no token-usage.jsonl. These workflows may not use --enable-api-proxy or may not make Claude API calls directly.

Per-Workflow Details

Daily Security Review and Threat Modeling

  • Run: §24002705293
  • Requests: 33 total
  • Provider: copilot (all 33 requests — via Copilot API proxy, not direct Anthropic)
  • Models: claude-sonnet-4.6 (33 reqs)
  • Tokens: 5,140K total (input: 2,620K, output: 20K, cache_read: 2,499K, cache_write: 0)
  • Cache hit rate: 48.8% (hits grow progressively across turns; first turn starts cold)
  • Cache write rate: 0.0% (no explicit Anthropic cache checkpoints set)
  • Avg I/O ratio: 512:1
  • Avg latency: 10,629ms/request
  • Estimated cost: $8.91 (at Anthropic list rates; actual billing via Copilot API)

Secret Digger (Claude)

  • Runs: 8 (§24001421625, §24002538112, §24003375166, §24004365799, §24005429223, §24006466570, §24007580596, §24008628142)
  • Requests: 24 total (3/run: 1×Haiku orchestrator + 2×Sonnet analysis)
  • Provider: anthropic (direct API)
  • Models: claude-haiku-4-5-20251001 (8 reqs), claude-sonnet-4-6 (16 reqs)
  • Tokens: 630K total (input: 1K, output: 12K, cache_read: 303K, cache_write: 314K)
  • Cache hit rate: 99.7% — excellent
  • Cache write rate: 50.8% — structurally sound (write on req 2, read back on req 3)
  • Avg I/O ratio: 930:1 (large context, small targeted outputs)
  • Avg latency: 10,201ms/request
  • Estimated cost: $1.44 total (~$0.18/run)

Per-run breakdown:

Run Tokens Cost
§24001421625 77.8K $0.17
§24002538112 79.9K $0.19
§24003375166 78.3K $0.18
§24004365799 78.6K $0.18
§24005429223 77.9K $0.17
§24006466570 80.4K $0.19
§24007580596 79.3K $0.18
§24008628142 78.2K $0.17
Workflows Without Token Data

The following workflows ran during this period but had no token-usage.jsonl in their agent-artifacts. They likely do not use --enable-api-proxy or make direct LLM API calls:

Workflow Runs Conclusion Notes
Agentic Maintenance 4 success No agent-artifacts at all
Firewall Issue Dispatcher 2 success No token-usage.jsonl in artifacts
Documentation Maintainer 1 success No token-usage.jsonl in artifacts
Smoke Claude 1 failure Different artifact format; no api-proxy data

Historical Trend

Date Secret Digger Runs Secret Digger Cost Daily Security Review Total Cost
2026-04-05 (today) 8 $1.44 (~$0.18/run) $8.91 (1 run, Copilot API) $10.35
2026-04-04 §1670 4 $0.74 (~$0.185/run) $0.74
2026-04-03 §1646 ~4 ~$0.74 ~$0.74

Secret Digger cost per run is stable at ~$0.18/run — no regression detected. The large jump in total cost today is due to the first appearance of Daily Security Review data, not any change in Secret Digger behavior.


Previous Report

§1670 — 📊 Claude Token Usage Report 2026-04-04

References:

Generated by Daily Claude Token Usage Analyzer ·

Metadata

Metadata

Assignees

No one assigned

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions