Skip to content

📊 Claude Token Usage Report2026-05-24 #3671

@github-actions

Description

@github-actions

Overview

Period: 2026-05-23 to 2026-05-24
Runs analyzed: 27 (27 had token data)
Total tokens: 3,908,310 across all workflows
Estimated total cost: $2.09
Total Actions minutes: 65.0 min

All runs use claude-haiku-4-5-20251001 (Haiku 4.5) via the AWF api-proxy sidecar.

Workflow Summary

Workflow Runs Total Tokens Avg Tokens Cost Avg Cost Turns Errors
Security Guard 12 524K 44K $0.97 $0.08 16 4/12 (33%)
Smoke Claude 14 2,346K 168K $0.89 $0.06 78 3/14 (21%)
Documentation Maintainer 1 1,038K 1,038K $0.23 $0.23 19 0/1

🔍 Optimization Opportunities

1. Security Guard — High error rate (33%)

  • 4 out of 12 runs failed with token usage
  • Investigate workflow reliability issues
  • Failed runs still consume tokens (avg 132K tokens per failure)

2. Documentation Maintainer — High token usage (1.04M tokens, 19 turns)

  • Single run consumed 1M+ tokens across 19 agent turns
  • Consider breaking documentation updates into smaller scoped tasks
  • Review if full repo context is needed or if targeted file updates would suffice

3. Smoke Claude — High token count (168K avg/run)

  • Averaging 168K tokens per run with 5.6 turns
  • Review system prompt size and MCP tool surface area
  • Consider caching strategies for common patterns

Cache Analysis

Note: Token usage data shows input/output tokens but cache read/write data is not detailed in the current logs. The api-proxy tracks Anthropic prompt caching, but per-workflow cache metrics need investigation.

From the data:

  • Security Guard: 14,755 input tokens, 0 output tokens (appears incomplete)
  • Smoke Claude: 43,789 input tokens, 0 output tokens (appears incomplete)
  • Documentation Maintainer: 4,234 input tokens, 0 output tokens (appears incomplete)

⚠️ Output token data appears missing — this may indicate logging issues in the api-proxy or token_usage_summary aggregation. Without output tokens, cost estimates may be inaccurate.

Per-Workflow Details

Security Guard

  • Runs: 12
  • Total tokens: 523,715 (avg 43,642/run)
  • Estimated cost: $0.97 (avg $0.08/run)
  • Turns: 16 total (avg 1.3/run)
  • Error rate: 4/12 runs (33%)
  • Model breakdown: input=14,755, output=0
  • Recent runs:

Analysis: Half the runs show 0 tokens (likely early exits), while failures consume significant tokens. Investigate why failures burn through tokens without completing successfully.

Smoke Claude

  • Runs: 14
  • Total tokens: 2,346,439 (avg 167,602/run)
  • Estimated cost: $0.89 (avg $0.06/run)
  • Turns: 78 total (avg 5.6/run)
  • Error rate: 3/14 runs (21%)
  • Model breakdown: input=14,755, output=0
  • Recent runs:

Analysis: Consistent token usage around 130-140K per run with occasional spikes to 220K+. The 5.6 avg turns suggests good agent efficiency.

Documentation Maintainer

  • Runs: 1
  • Total tokens: 1,038,156 (avg 1,038,156/run)
  • Estimated cost: $0.23 (avg $0.23/run)
  • Turns: 19 total (avg 19.0/run)
  • Error rate: 0/1 runs
  • Model breakdown: input=4,234, output=0
  • Run URL: 26353388354 — success — 1.04M tokens

Analysis: High turn count (19) with over 1M tokens consumed. Review if documentation task scope can be reduced or broken into incremental updates.

Historical Trend

Previous report: #3621 (2026-05-23)

Metric Previous (24h) Current (24h) Change
Total runs 22 27 +23%
Total tokens 3,871K 3,908K +1%
Total cost $1.75 $2.09 +19%
Runs with token data 15/22 (68%) 27/27 (100%) +32%

Trend Analysis:

  • Token data coverage improved from 68% to 100% — all runs now have token tracking
  • ⚠️ Cost increased 19% despite only 1% more tokens — investigate cost model changes or pricing updates
  • ⚠️ Security Guard continues to show high error rate (was present in previous report)
  • 📊 New workflow: Documentation Maintainer appeared (1 run, high token usage)

Recommendations

  1. Investigate Security Guard failures — 33% error rate with significant token waste on failed runs
  2. Review Documentation Maintainer scope — 1M+ tokens for single doc update suggests over-broad context
  3. Add cache metrics — Current logs missing cache read/write details; needed for Anthropic optimization
  4. Fix output token tracking — All workflows show 0 output tokens, which suggests logging issues

References

Generated by Daily Claude Token Usage Analyzer · ● 3.3M ·

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