📊 Claude Token Usage Report2026-04-20

[github-actions]

Overview

Period: 2026-04-19 to 2026-04-20
Runs analyzed: 40 (32 had token data)
Total tokens: 6,178K (effective: 1,906K after cache discounting)
Estimated total cost: $3.66
Total Actions minutes: 140 min
Total turns: 212

---

Workflow Summary

Workflow	Runs	With Data	Total Tokens	Avg Tokens	Total Cost	Avg Cost	Avg Turns	Errors	Model
Security Guard	16	9	2,126K	236K	$2.10	$0.23	5.6	1	claude-sonnet-4-6
Smoke Claude	24	23	4,052K	176K	$1.56	$0.07	7.0	2	claude-haiku-4-5

---

🔍 Optimization Opportunities

1. Security Guard — $0.23/run avg on claude-sonnet-4-6

   • Avg 236K tokens/run exceeds the 100K threshold — review system prompt size and MCP tool surface
   • 7 of 16 runs had no token data (missing api-proxy or token tracking not active)
   • Cache efficiency looks good: 1.75M cache reads vs 356K cache writes — prompt prefix is being reused well
   • 1 run ended with an error; monitor reliability

2. Smoke Claude — 24 runs on claude-haiku-4-5 (cheaper model, good choice for smoke tests)

   • Cache reads (3.15M) vs cache writes (865K) ratio ~3.6x — healthy cache reuse
   • 1 run had no token data; 2 errors recorded — minor, but worth checking

---

Per-Workflow Details

Security Guard

• Runs: 16 total, 9 with token data, 7 missing data
• Total tokens: 2,126K (avg 236K/run with data)
• Estimated cost: $2.10 (avg $0.23/run with data)
• Turns: 50 total (avg 5.6/run)
• Model: claude-sonnet-4-6 (Anthropic)
• Cache analysis: 1,754K cache reads / 356K cache writes — cache efficiency ~83% read reuse
• Error rate: 1/9 instrumented runs (11%)
• Sample runs: 24687095846, 24684649486, 24684043315

Smoke Claude

• Runs: 24 total, 23 with token data, 1 missing data
• Total tokens: 4,052K (avg 176K/run with data)
• Estimated cost: $1.56 (avg $0.07/run with data)
• Turns: 162 total (avg 7.0/run)
• Model: claude-haiku-4-5-20251001 (Anthropic — cost-optimized)
• Cache analysis: 3,154K cache reads / 865K cache writes — cache efficiency ~78% read reuse
• Error rate: 2/23 instrumented runs (9%)
• Sample runs: 24687095878, 24684649496, 24684043333

Runs Without Token Data

8 runs had token_usage = null — these runs may lack the api-proxy sidecar or ran before token tracking was enabled:

• Security Guard: 7 runs with no token data (likely short runs or runs where the agent didn't make LLM calls)
• Smoke Claude: 1 run with no token data

To instrument these, ensure --enable-api-proxy is set in the workflow and that AWF is routing Claude API calls through the sidecar proxy.

---

Historical Trend

This is the first Claude token usage report for this repository. Future reports will show token and cost trends over time.

---

References:

• Workflow Run: §24687710872

Generated by Daily Claude Token Usage Analyzer · ● 197K · ◷