📊 Claude Token Usage Report2026-06-12

[github-actions]

Overview

Period: 2026-06-11 18:02 UTC to 2026-06-12 02:14 UTC (~8.2 hours)
Runs analyzed: 50 (39 had token data — 78% coverage ↑ from 60% yesterday; 18 had AIC data — 36%)
Total tokens: 6,243K across all workflows
Total AIC: 648.94 (partial — 18/50 runs have AIC data; recent runs pending)
Total Actions minutes: 409 min | Total turns: 114 | Total errors: 13

⚠️ Smoke Claude failure rate worsened — 12/16 runs failed (75%), up from 56% yesterday. All failing runs hit exactly 4 turns with ~191–193K tokens. This persistent pattern needs urgent investigation.

✅ Contribution Check fully recovered — 0/17 failures today vs 4/16 yesterday. Token usage stable at ~122K/run.

✅ Security Guard improved — 1/17 failures (6%) vs 4/17 yesterday. 11/17 runs exited early (no tokens = guard correctly noop-ing on non-security-relevant PRs).

i️ Documentation Maintainer — appeared yesterday (1 run, failed). No runs today.

---

Workflow Summary

Workflow	Runs	w/ Data	Total Tokens	Avg Tokens	AIC (partial)	Avg AIC	Turns	Avg Turns	Failures
Smoke Claude	16	16	3,006K	188K	260.19	16.26	60	3.8	12/16 ⚠️
Contribution Check	17	17	2,071K	122K	235.28	13.84	34	2.0	0/17 ✅
Security Guard	17	6	1,166K	194K†	153.47	9.03	20	1.2	1/17
Total	50	39	6,243K	—	648.94	—	114	—	13/50

† Security Guard avg over 6 data-bearing runs; 11/17 exits were early-exit noops (0 tokens — expected behavior).

---

🔍 Optimization Opportunities

1. Smoke Claude — 75% failure rate (>> 30% threshold), avg 188K tokens/run (> 100K threshold)

   • 11 of 12 failing runs hit exactly 4 turns with ~191–193K tokens — highly deterministic failure signature
   • One run succeeded with 4 turns ([aw] Security Guard failed #4459), ruling out a hard turn limit as the cause
   • Pattern persists across multiple PR branches by the same actor (lpcox) and has worsened day-over-day
   • Recommendation: Examine what happens at turn 4 — look at the agent's last tool call or error message. Check if a specific workflow step or tool is timing out or returning a blocking error.

2. Contribution Check — avg 122K tokens/run (> 100K threshold), 13/17 runs have missing_data_count=1

   • Every run triggering safeoutputs missing_data suggests the workflow depends on data that is regularly unavailable
   • Recommendation: Investigate which missing_data call is firing — if it's expected (e.g., no prior issue found), consider suppressing it or treating as a noop signal.

3. Security Guard (active runs) — avg 194K tokens/run over 6 data-bearing runs (> 100K threshold)

   • Runs that do engage use almost as many tokens as Smoke Claude despite a narrower scope
   • Two runs used 261K–274K tokens (5 turns each)
   • Recommendation: Review high-turn Security Guard runs to see if they're doing redundant analysis.

---

Per-Workflow Details

Smoke Claude

• Runs: 16 (#4450 #4451 #4453 #4455 #4456 #4457 #4458 #4459 #4460 #4461 #4462 #4463 #4464 #4465 #4466 #4467)
• Total tokens: 3,006K (avg 188K/run)
• AIC: 260.19 (6 runs with data, avg 16.26/run)
• Turns: 60 total (avg 3.8/run)
• Action minutes: 215 min
• Failure pattern:
  • ❌ 4 turns, ~191–193K tokens (11 runs)
  • ✅ 4 turns, 192K tokens (1 run — [aw] Security Guard failed #4459)
  • ✅ 3 turns, ~192K tokens (2 runs — [Coverage Report] Test Coverage Report — 2026-06-06 #4460, [aw] Build Test Suite failed #4462)
  • ✅ 2 turns, 125K tokens (1 run — [aw] Smoke Copilot BYOK AOAI (Entra) failed #4464)
• Error rate: 12/16 runs failed (75%) ⚠️

Contribution Check

• Runs: 17 (all successful)
• Total tokens: 2,071K (avg 122K/run) — extremely consistent (121K–123K range)
• AIC: 235.28 (8 runs with data, avg 13.84/run)
• Turns: 34 total (avg 2.0/run — consistent 2-turn pattern)
• Action minutes: 107 min
• Missing data: 13/17 runs called safeoutputs missing_data once
• Error rate: 0/17 ✅

Security Guard

• Runs: 17 total; 6 with token data, 11 early-exit (0 tokens)
• Total tokens (data-bearing runs): 1,166K (avg 194K over 6 runs; avg 69K over all 17)
• AIC: 153.47 (4 runs with data, avg 9.03/run)
• Turns: 20 total (avg 1.2/run incl. zero-turn noops)
• Action minutes: 87 min
• Token breakdown:

Run	Tokens	Turns	Status
#3985	264K	5	✅ success
#3986	122K	2	✅ success
#3987	123K	2	✅ success
#3990	122K	2	✅ success
#3991	262K	4	✅ success
#4000	274K	5	❌ failure
11 runs	0	0	✅ early-exit noop

• Error rate: 1/17 (6%)

Runs Without Token Data

11 Security Guard runs exited early with no token usage — this is expected behavior when the guard determines no security-relevant changes are present. These runs are not instrumentation gaps.

Run #	Created	Conclusion
#3984	2026-06-11T18:08	✅ success
#3988	2026-06-11T18:32	✅ success
#3989	2026-06-11T18:37	✅ success
#3992	2026-06-11T22:09	✅ success
#3993	2026-06-12T00:44	✅ success
#3994	2026-06-12T00:55	✅ success
#3995	2026-06-12T01:03	✅ success
#3996	2026-06-12T01:18	✅ success
#3997	2026-06-12T01:29	✅ success
#3998	2026-06-12T01:33	✅ success
#3999	2026-06-12T01:45	✅ success

---

Historical Trend

Metric	2026-06-11	2026-06-12	Change
Total tokens	4,967K	6,243K	+26% ↑
Total AIC (partial)	1,445.29	648.94	−55%*
Total failures	18/50	13/50	−5 ✅
Smoke Claude failures	9/16 (56%)	12/16 (75%)	+19% ⚠️
Contribution Check failures	4/16 (25%)	0/17 (0%)	−25% ✅
Security Guard failures	4/17 (24%)	1/17 (6%)	−18% ✅
Token coverage	30/50 (60%)	39/50 (78%)	+18% ✅

* AIC comparison unreliable: today's AIC is from 18/50 runs (recent runs lack AIC data), vs yesterday's coverage. Overall cost trend unclear until full AIC data is available.

Key observations:

• Smoke Claude degradation is accelerating (4th consecutive day with high failure rate; severity increasing)
• Contribution Check and Security Guard have both recovered significantly
• Token usage increased despite fewer overall failures — likely driven by more Smoke Claude runs

Previous Report

§27407634330 | 📊 Claude Token Usage Report 2026-06-11 (#4739)

References:

• §27407634330 — this workflow run
• #4739 — previous report (2026-06-11)
• #4667 — report from 2026-06-10

Generated by Daily Claude Token Usage Analyzer · ◷

[github-actions]

This issue is being closed as outdated. A newer issue has been created: #4862

View newer issue

---

This action was performed automatically by the Daily Claude Token Usage Analyzer workflow.