📊 Daily Copilot Token Consumption Report — 2026-03-23

[github-actions[bot]]

Executive Summary

Over the current reporting period, 44 Copilot-powered agentic workflows consumed 97,580,594 tokens across 113 workflow runs with 1,926 agent turns total. No cost data is available in this period's logs. The average token consumption was 863,545 tokens per run, reflecting a mix of lightweight triage workflows and intensive deep-analysis agents.

Key Highlights

Metric	Value
🔢 Total Tokens	97,580,594 (97.6M)
🔄 Total Runs	113 across 44 unique workflows
🔁 Total Agent Turns	1,926
⚡ Avg Tokens/Run	863,545
🔝 Most Tokens (single run)	Code Simplifier — 8,975,807 tokens (105 turns)
🏃 Most Active Workflow	Issue Monster — 41 runs
⚠️ Error Count	9 errors across all runs
📊 Historical Days Tracked	7 days

---

📈 Token Usage & Run Count Trends

Daily token consumption visualized against workflow run counts. Today's run (2026-03-23) represents 97.6M tokens across 113 runs — a significant volume driven by high-turn deep-analysis workflows alongside frequent Issue Monster executions.

---

⚡ Token Efficiency: Avg Tokens per Run

Average tokens per run over time. High single-run agents (Code Simplifier, Dead Code Removal, Glossary Maintainer) significantly inflate this metric. Excluding single-run deep agents, the average drops to ~450K tokens/run, consistent with multi-turn analytical workflows.

---

🏆 Top Workflows by Token Consumption

Top 10 Most Token-Intensive Workflows

Rank	Workflow	Total Tokens	Runs	Avg Tokens/Run	Avg Turns
1	Issue Monster	20,693,324	41	504,715	13.9
2	Code Simplifier	8,975,807	1	8,975,807	105.0
3	Glossary Maintainer	6,974,174	1	6,974,174	73.0
4	Dead Code Removal Agent	6,080,894	1	6,080,894	80.0
5	Daily Syntax Error Quality Check	4,061,694	1	4,061,694	73.0
6	Auto-Triage Issues	3,501,342	10	350,134	8.9
7	Dev	3,328,043	6	554,674	12.7
8	GitHub Remote MCP Authentication Test	3,289,458	1	3,289,458	85.0
9	Workflow Skill Extractor	3,054,853	1	3,054,853	34.0
10	Contribution Check	2,994,957	6	499,160	14.3

All 44 Workflows — Complete Statistics

Workflow	Total Tokens	Runs	Avg Tokens/Run	Avg Turns	Avg Duration
Issue Monster	20,693,324	41	504,715	13.9	4.9m
Code Simplifier	8,975,807	1	8,975,807	105.0	18.1m
Glossary Maintainer	6,974,174	1	6,974,174	73.0	14.1m
Dead Code Removal Agent	6,080,894	1	6,080,894	80.0	21.5m
Daily Syntax Error Quality Check	4,061,694	1	4,061,694	73.0	10.7m
Auto-Triage Issues	3,501,342	10	350,134	8.9	4.3m
Dev	3,328,043	6	554,674	12.7	8.1m
GitHub Remote MCP Authentication Test	3,289,458	1	3,289,458	85.0	7.3m
Workflow Skill Extractor	3,054,853	1	3,054,853	34.0	13.6m
Contribution Check	2,994,957	6	499,160	14.3	4.4m
Copilot CLI Deep Research Agent	2,982,919	1	2,982,919	39.0	10.1m
Layout Specification Maintainer	2,594,730	1	2,594,730	41.0	8.5m
jsweep - JavaScript Unbloater	2,519,297	1	2,519,297	39.0	11.9m
Smoke Multi PR	2,339,215	1	2,339,215	53.0	8.7m
PR Triage Agent	2,160,626	4	540,156	12.0	4.3m
Copilot PR Conversation NLP Analysis	1,898,176	1	1,898,176	33.0	21.4m
Daily CLI Performance Agent	1,882,045	1	1,882,045	33.0	9.9m
Daily Compiler Quality Check	1,778,656	1	1,778,656	25.0	8.7m
Daily Community Attribution Updater	1,424,736	1	1,424,736	23.0	7.2m
Daily News	1,375,609	1	1,375,609	26.0	9.6m
Dependabot Dependency Checker	1,264,615	1	1,264,615	23.0	9.8m
Workflow Health Manager - Meta-Orchestrator	1,209,834	1	1,209,834	21.0	7.2m
Organization Health Report	1,183,393	1	1,183,393	22.0	8.0m
Daily Malicious Code Scan Agent	1,054,248	1	1,054,248	20.0	5.7m
Daily Secrets Analysis Agent	904,014	1	904,014	19.0	5.9m
Agent Container Smoke Test	788,829	6	131,472	3.7	4.0m
Daily Regulatory Report Generator	740,108	1	740,108	12.0	7.1m
Sub-Issue Closer	692,951	1	692,951	17.0	5.0m
Copilot PR Prompt Pattern Analysis	684,343	1	684,343	15.0	6.7m
Terminal Stylist	682,363	1	682,363	14.0	5.6m
Daily Safe Output Integrator	626,408	1	626,408	12.0	4.6m
Daily MCP Tool Concurrency Analysis	623,337	1	623,337	11.0	5.8m
Architecture Diagram Generator	537,910	1	537,910	12.0	6.5m
GPL Dependency Cleaner (gpclean)	513,329	1	513,329	12.0	4.5m
Daily Project Performance Summary Generator (Using MCP Scripts)	423,254	1	423,254	9.0	6.3m
Draft PR Cleanup	416,080	1	416,080	11.0	3.9m
Daily Workflow Updater	338,769	1	338,769	9.0	5.3m
Smoke Update Cross-Repo PR	253,119	2	126,560	5.0	4.1m
Smoke Create Cross-Repo PR	246,255	2	123,128	5.0	3.9m
Dependabot Burner	214,351	1	214,351	6.0	3.6m
Constraint Solving — Problem of the Day	195,024	1	195,024	5.0	4.5m
The Great Escapi	77,505	1	77,505	2.0	3.7m
Documentation Noob Tester	0	1	0	0.0	21.2m
Daily Copilot Token Consumption Report	0	1	0	0.0	33s

---

💡 Insights & Recommendations

High-Token Single-Run Agents (Top Concern)

Four workflows consumed 3–9M tokens in a single run due to extremely high turn counts (73–105 turns):

1. Code Simplifier — 8.97M tokens, 105 turns, 18.1m duration

   • Observation: Highest per-run consumption by a wide margin. 105 turns suggest deeply iterative code analysis with large context windows.
   • Recommendation: Scope analysis to specific packages or files per run rather than the full codebase. Consider adding early-exit conditions.

2. Dead Code Removal Agent — 6.08M tokens, 80 turns, 21.5m duration

   • Observation: 80 turns with longest average duration (21.5m). Likely doing extensive cross-reference analysis.
   • Recommendation: Run incrementally on recently-changed files only. Use a diff-based trigger (push) instead of full scan.

3. Glossary Maintainer — 6.97M tokens, 73 turns, 14.1m

   • Observation: 73 turns suggests repeated context loading for glossary lookups and updates.
   • Recommendation: Cache glossary state between runs to avoid re-scanning the entire codebase each time.

4. GitHub Remote MCP Authentication Test — 3.29M tokens, 85 turns

   • Observation: 85 turns for what should be a test workflow is unexpectedly high.
   • Recommendation: Review this workflow for runaway loops or excessive verification steps.

High-Volume Recurring Workflows

5. Issue Monster — 20.7M tokens (21.2% of total), 41 runs
   • Observation: The rejig docs #1 consumer by total tokens despite reasonable per-run cost (~504K avg). High frequency (41 runs) is the driver.
   • Recommendation: Evaluate if 41 runs/period is necessary. Consider batching issues or adding deduplication to avoid redundant processing.

Optimization Opportunities

Opportunity	Affected Workflows	Potential Token Savings	Action
Scope single-run deep agents	Code Simplifier, Dead Code Removal, Glossary Maintainer	~15M tokens	Limit analysis to changed files or specific packages per run
Reduce Issue Monster run frequency	Issue Monster	~5–10M tokens	Batch issues; run every 30m instead of on each event
Cap max turns	Code Simplifier (105), GitHub MCP Test (85), Dead Code (80)	~8M tokens	Add max_turns limit of 40–50 per run
Cache context between runs	Glossary Maintainer, Workflow Skill Extractor	~4M tokens	Use cache-memory to persist index/state between runs
Parallelize smoke tests	Smoke Multi PR (53 turns)	~1M tokens	Split into smaller focused tests

Estimated total savings potential: ~28–38M tokens per similar period

Efficiency Insights

• Most efficient recurring workflow: Agent Container Smoke Test — 131K avg tokens/run, 3.7 avg turns, 4.0m duration. Good benchmark for lightweight test workflows.
• Best turns-to-depth ratio: Auto-Triage Issues — 8.9 avg turns, 350K avg tokens. Well-scoped per run.
• Documentation Noob Tester: 0 tokens recorded despite 21.2m runtime — likely a token-reporting issue or non-Copilot execution path.
• High-turn outliers (>40 turns): Code Simplifier (105), GitHub Remote MCP Auth Test (85), Dead Code Removal (80), Glossary Maintainer (73), Daily Syntax Error Quality Check (73), Smoke Multi PR (53), Layout Specification Maintainer (41), Copilot CLI Deep Research (39), jsweep (39).

Historical Trend Comparison

Date	Total Tokens	Runs	Avg Tokens/Run
2026-01-15	94,899,840	178	533,146
2026-01-22	101,969,175	239	426,649
2026-01-28	200,676,032	281	714,149
2026-02-11	237,834,034	378	629,191
2026-02-19	119,326,605	83	1,437,669
2026-02-20	166,507,574	195	854,398
2026-03-23	97,580,594	113	863,545

Key observations:

• Today's total (97.6M) is below the historical average of ~145M tokens/day, suggesting either reduced workflow activity or improved efficiency.
• The run count (113) is significantly lower than the Feb 11 peak (378 runs), which may reflect workflow consolidation or scheduling changes.
• The avg tokens/run (863K) is above the Jan averages (~450K–530K) but inline with Feb, indicating deeper/more complex workflows are now running even if fewer runs occur.

Methodology & Data Quality

Methodology

• Data Source: Pre-downloaded GitHub Actions workflow run artifacts (/tmp/gh-aw/copilot-logs.json)
• Engine Filter: Copilot agent workflows only
• Analysis Date: 2026-03-23T11:35 UTC
• Historical Data: 7 days tracked (stored in memory/token-metrics/history.jsonl)
• Duration Parsing: String durations (4.5m, 33.0s) converted to seconds for averaging

Data Quality Notes

• 5 runs had no token_usage field (excluded from per-workflow token totals): includes the current in-progress report run and Documentation Noob Tester
• Cost data: Not present in this period's log export — historical records show cost was tracked in prior periods
• 1 run was in_progress at time of data collection (Daily Copilot Token Consumption Report itself)
• 9 errors recorded across runs; no missing-tool events; 5 missing-data events

---

Recommendations Summary

Priority	Action	Impact
🔴 High	Cap max turns for deep-analysis agents (Code Simplifier, Dead Code, Glossary) at 40–50 turns	-15M tokens
🔴 High	Scope Code Simplifier and Dead Code Removal to diff/changed files only	-8M tokens
🟡 Medium	Reduce Issue Monster run frequency or add batching	-5–10M tokens
🟡 Medium	Cache context between Glossary Maintainer and Workflow Skill Extractor runs	-4M tokens
🟢 Low	Investigate GitHub Remote MCP Authentication Test's 85-turn count	-2M tokens

---

References:

• §23434633325 — Current report run
• §23432345784 — Glossary Maintainer (6.97M tokens)
• §23432366934 — Issue Monster sample run

AI generated by Daily Copilot Token Consumption Report · history

• expires on Mar 26, 2026, 11:39 AM UTC

[github-actions[bot]]

This discussion has been marked as outdated by Daily Copilot Token Consumption Report.

A newer discussion is available at Discussion #22661.