Repository navigation
Closed as not planned
Description
Overview
Period: 2026-06-03 to 2026-06-04
Runs analyzed: 50 (39 had token data)
Total tokens: 11,391K across all workflows
Estimated total cost: $7.82
Total Actions minutes: 174 min
Workflow Summary
| Workflow | Runs | Runs w/ Data | Total Tokens | Avg Tokens/Run | Total Cost | Avg Cost/Run | Total Turns | Avg Turns | Errors |
|---|---|---|---|---|---|---|---|---|---|
| Security Guard | 23 | 12 | 5,088K | 424K | $5.36 | $0.23 | 74 | 3.2 | 4 |
| Smoke Claude | 26 | 26 | 6,080K | 234K | $2.36 | $0.09 | 99 | 3.8 | 0 |
| Documentation Maintainer | 1 | 1 | 223K | 223K | $0.10 | $0.10 | 5 | 5.0 | 1 |
🔍 Optimization Opportunities
-
Security Guard — $0.23/run avg, 424K avg tokens/run (>100K threshold), 17% error rate
- Each Security Guard run uses ~424K tokens on average — well above the 100K threshold. With 23 runs in this period it's the largest cost driver at $5.36 total.
- 4 out of 12 instrumented runs failed (33% error rate — above the 30% threshold). Investigate the failure cause: repeated failures mean paying for failed work.
- 11 of 23 runs have no token data, suggesting the api-proxy may not be enabled consistently across all trigger conditions.
- Recommendation: Review the Security Guard prompt and tool surface to reduce per-turn token load. Consider splitting the task or adding early-exit conditions for non-security-relevant inputs.
-
Security Guard — 11 runs missing token data
- These runs show
token_usage = 0despite beingsuccessconclusions. This likely means the api-proxy sidecar was not active for those runs (e.g., triggered via a different event or branch path that skips AWF). - Recommendation: Audit the workflow trigger conditions. Ensure all Security Guard runs flow through AWF with
--enable-api-proxyto capture full token telemetry.
- These runs show
-
Smoke Claude — high run frequency (26 runs/day)
- Smoke Claude is the highest-volume workflow. While per-run cost is low ($0.09), the volume drives $2.36/day. All 26 runs succeeded with no errors.
- Recommendation: If smoke tests run on every push, consider rate-limiting to once per branch or deduplicate on identical SHAs.
Per-Workflow Details
Security Guard
- Runs: 23 total — 26922288525 26919966996 26919371105 26919029541 26917205554 26916812491 26915260910 26903276727 26902991499 26900286614 26899670906 26896741400 (+11 runs without token data)
- Total tokens: 5,088K (avg 424K/run for instrumented runs)
- Estimated cost: $5.36 (avg $0.23/run)
- Turns: 74 total (avg 3.2/run across all; avg ~6/run for instrumented runs)
- Cache analysis:
token_usage_summarynot available — api-proxy may use a passthrough model without cache tracking - Error rate: 4/23 runs had errors (17%); 4 failures in the
failureconclusion runs
Smoke Claude
- Runs: 26 — all successful, all with token data
- Total tokens: 6,080K (avg 234K/run)
- Estimated cost: $2.36 (avg $0.09/run)
- Turns: 99 total (avg 3.8/run)
- Token range: 183K–307K per run (tight distribution — consistent workload)
- Cache analysis:
token_usage_summarynot available - Error rate: 0/26 runs
Documentation Maintainer
- Runs: 1
- Total tokens: 223K
- Estimated cost: $0.10
- Turns: 5
- Conclusion: failure (1 error)
- Run: 26934274043
Runs Without Token Data (11 runs)
The following Security Guard runs reported token_usage = 0. These runs completed successfully but produced no token telemetry, suggesting the api-proxy was not active:
| Run | Conclusion |
|---|---|
| 26897356098 | success |
| 26896994136 | success |
| 26896922950 | success |
| 26896795991 | success |
| 26896756028 | success |
| 26896713057 | success |
| 26892609541 | success |
| 26892576775 | success |
| 26890586426 | success |
| 26890395773 | success |
| 26890367458 | success |
Historical Trend
This is the first Claude token usage report — no prior baseline to compare against.
References:
- §26943456630 — this analysis run
Generated by Daily Claude Token Usage Analyzer · sonnet46 1M · ◷
Activity
github-actions commentedon Jun 5, 2026
This issue is being closed as outdated. A newer issue has been created: #4368
View newer issue
This action was performed automatically by the
Daily Claude Token Usage Analyzerworkflow.