📊 Copilot Token Usage Report2026-04-04

[github-actions]

Overview

Period: 2026-04-04T07:26Z to 2026-04-04T10:18Z
Runs analyzed: 4 Secret Digger (Copilot) runs (all had token data); 4 other Copilot-engine runs (no artifacts)
Total tokens: 752K across instrumented workflows
Estimated total cost: $1.16 (Copilot endpoint rates: $2.50/M input, $10/M output)

⚠️ Recurring failure pattern: The retry-loop failure in Secret Digger (Copilot) seen in the previous report has recurred. Run §23975973080 shows the identical stuck pattern — requests 2–7 all return output_tokens = 2 at identical input_tokens = 35,237 — costing 3.5× a normal run ($0.62 vs. $0.18). This is now a confirmed recurring issue.

📉 Documentation Maintainer absent: No doc-maintainer runs observed this period (previously $3.91/run). Total cost is significantly lower as a result.

Workflow Summary

Workflow	Runs	Requests	Total Tokens	Est. Cost	Cache Rate	I/O Ratio	Top Model
Secret Digger (Copilot)	4	15	752K	$1.16	33.6%	498:1	sonnet-4.6
Total	4	15	752K	$1.16	33.6%	498:1

🔍 Optimization Opportunities

1. Secret Digger (Copilot) — recurring retry loop (confirmed bug)

   • Second occurrence in two consecutive daily reports
   • Run §23975973080 (failure): 7 requests vs. normal 2
   • Requests 2–7: all output_tokens = 2 and identical input_tokens = 35,237 — a stuck loop producing no progress
   • Wasted cost: $0.62 vs. $0.18 normal (3.5× overhead); by request 3 cache is fully warm (cache_read = 35,236 = input_tokens) yet output doesn't improve
   • Recommendation: Add a max-iterations guard or detect when consecutive outputs are identical/minimal; escalate to workflow maintainer

2. Cache hit rate plateau at ~30% for normal success runs

   • All 3 success runs show exactly 29.7% cache hit rate — consistent but not improving
   • Request 1 of each run starts cold (0 cache reads), request 2 hits ~29K cached tokens
   • The first request always incurs full input cost; caching only kicks in from request 2+
   • Recommendation: Evaluate whether the static system prompt prefix can be pre-warmed, or structure prompts to maximize the cached prefix length

3. Cache write tokens remain 0 (provider limitation)

   • Consistent across all runs — the Copilot inference endpoint doesn't report cache_write_tokens separately
   • Cost estimates may be slightly underestimated (missing cache write billing)
   • Status: Known limitation, no action needed

Per-Run Details

Secret Digger (Copilot)

Run	Conclusion	Requests	Input	Output	Cache Read	Total	Cache Rate	Cost
§23976844295	✅ success	2	69,845	277	29,501	99,623	29.7%	$0.18
§23975973080	❌ failure	7	246,189	417	205,677	452,283	45.5%	$0.62
§23975039884	✅ success	2	69,977	417	29,495	99,889	29.7%	$0.18
§23974187599	✅ success	2	69,980	445	29,499	99,924	29.7%	$0.18

Normal run profile (avg of 3 success runs):

• Requests: 2 (request 1 cold, request 2 warm)
• Tokens: ~99.8K (70K input, ~380 output, ~29.5K cache_read)
• Avg latency: ~6,300ms/request
• Cache hit rate: 29.7% (second request only)
• Cost per run: ~$0.18

Failure run analysis (7 requests):

• Request 1: 34,767 input, 405 output, 0 cache_read — normal start
• Requests 2–7: 35,237 input, 2 output, 29,497–35,236 cache_read — stuck loop
• Cache fully warm by request 3 but model output frozen at 2 tokens
• Additional 5 wasted calls: ~$0.44 overhead

Workflows Without Token Data

The following Copilot-engine workflows ran but produced no agent-artifacts with token data:

Workflow	Run	Conclusion	Reason
Smoke Copilot	§23974551292	❌ failure	No agent-artifacts uploaded (likely failed before AWF started)
Smoke Services	§23974551273	❌ failure	No agent-artifacts uploaded
Build Test	§23974551276	❌ failure	No agent-artifacts uploaded
Agentic Maintenance	§23975583864	✅ success	No agent-artifacts (may not use --enable-api-proxy)

Note: The three failures (Smoke Copilot, Smoke Services, Build Test) all triggered from the same PR push at 07:48Z and appear to have failed at the AWF/infrastructure level rather than consuming LLM tokens.

Historical Trend

Period	Workflow	Runs	Total Tokens	Cost	Cache Rate	Notes
2026-04-04	Secret Digger (Copilot)	4	752K	$1.16	33.6%	1 failure (retry loop)
2026-04-03	Secret Digger (Copilot)	5	850K	$1.33	38.1%	1 failure (retry loop)
2026-04-03	Documentation Maintainer	1	3,003K	$3.91	48.7%	First appearance
2026-04-03	Total	6	3,853K	$5.24	46.0%

Trend observation: Without the Documentation Maintainer running today, total cost dropped from $5.24 to $1.16 (-78%). The Secret Digger (Copilot) per-run cost is stable at ~$0.18 for successful runs. The retry-loop failure pattern has now occurred in two consecutive daily reports — this warrants immediate attention.

Previous Report

📊 Copilot Token Usage Report 2026-04-03

References:

• §23976844295 — Latest Secret Digger (Copilot) success
• §23975973080 — Failure run with retry loop
• §23976915564 — This report's workflow run

Generated by Daily Copilot Token Usage Analyzer · ◷