📊 Daily Copilot Token Consumption Report — 2026-04-01

[github-actions[bot]]

Executive Summary

Over the reporting period, 133 Copilot-powered workflow runs across 65 unique workflows consumed 94,734,697 tokens (~94.7M), completing 1,865 agent turns in 1,199 action minutes (~20 hours). Cost data is not available in this dataset (all $0.00). Compared to the February 2026 peak (237.8M tokens, 378 runs), today's consumption is 60% lower, returning to early-January baseline levels — a healthy sign of workflow optimization.

Key Highlights

• 🏆 Highest single-run cost: Daily Syntax Error Quality Check — 11.4M tokens, 168 turns (failed with 1 error)
• 🔁 Most active workflow: Issue Monster — 20 runs, averaging 173K tokens/run
• ⚡ Most efficient high-volume workflow: Agent Container Smoke Test — 10 runs at only 142K tokens avg
• 🤖 Agentic efficiency flags: 65 resource_heavy_for_domain, 41 model_downgrade_available, 68 partially_reducible assessments across all runs
• ❌ Errors: 16 errors across 13 workflows — Smoke Update Cross-Repo PR (3 errors) and Smoke Create Cross-Repo PR (2 errors) are repeat offenders

---

📈 Token Usage & Historical Trends

Token consumption over the tracked history shows a clear peak in February 2026 (up to 237.8M tokens on 2026-02-11) followed by a sustained decline back to ~95M tokens today. Run count has similarly decreased from 378 to 133 — roughly 65% fewer runs than the peak period.

Historical Daily Totals

Date	Total Tokens	Runs	Notes
2026-01-15	94,899,840	178	Baseline
2026-01-22	101,969,175	239	↑ +7%
2026-01-28	200,676,032	281	↑ +97% spike
2026-02-11	237,834,034	378	🔴 Peak consumption
2026-02-19	119,326,605	83	↓ -50% drop
2026-02-20	166,507,574	195	Brief recovery
2026-04-01	94,734,697	133	Today — at baseline

---

🏆 Top 10 Workflows by Token Consumption

Rank	Workflow	Total Tokens	Runs	Avg Tok/Run	Turns	Avg Turns/Run
1	Daily Syntax Error Quality Check ❌	11,378,378	1	11,378,378	168	168
2	Contribution Check	6,734,002	6	1,122,334	204	34
3	Daily Community Attribution Updater	4,825,564	1	4,825,564	97	97
4	Daily Testify Uber Super Expert	4,492,441	1	4,492,441	62	62
5	Auto-Triage Issues	4,412,438	11	401,131	104	9.5
6	Documentation Noob Tester ❌	3,742,876	1	3,742,876	69	69
7	Dead Code Removal Agent	3,670,585	1	3,670,585	52	52
8	Issue Monster	3,460,298	20	173,015	100	5
9	Daily Compiler Quality Check	3,277,331	1	3,277,331	46	46
10	Code Simplifier	2,821,647	1	2,821,647	45	45

---

🔁 Most Active Workflows

The most frequently-run workflows with token efficiency context:

Workflow	Runs	Avg Tokens/Run	Total Tokens
Issue Monster	20	173,015	3,460,298
Smoke Copilot	11	0	0
Auto-Triage Issues	11	401,131	4,412,438
Agent Container Smoke Test	10	141,934	1,419,341
Contribution Check	6	1,122,334	6,734,002
Q	6	0	0

---

💡 Insights & Recommendations

High-Cost Workflows

1. Daily Syntax Error Quality Check — 11.4M tokens, 168 turns, failed

   • This is the single most expensive run in the period and it failed. 168 turns suggests the agent looped or struggled to complete.
   • Recommendation: Add early exit conditions and a turn budget cap (~40 turns). Investigate the missing tool (missing_tool_count: 1). Consider a model downgrade to gpt-4.1-mini for syntax-oriented tasks.

2. Contribution Check — 6.7M tokens across 6 runs (avg 1.1M/run)

   • 5 of 6 runs flagged as resource_heavy_for_domain. At 34 avg turns per run this is high for a PR review workflow.
   • Recommendation: Pre-fetch PR diffs as a deterministic step before the agent runs to reduce data-gathering turns. Estimated ~50% turn reduction possible.

3. Daily Community Attribution Updater — 4.8M tokens, 97 turns, single run

   • Flagged as resource_heavy_for_domain with poor_agentic_control. 97 turns is very high for a maintenance task.
   • Recommendation: Break this into smaller, focused sub-tasks. Move GitHub API data-gathering to a deterministic frontmatter step.

Agentic Assessment Summary (All Runs)

Assessment Type	Count	Severity
partially_reducible	68	Low — ~50% of turns can move to deterministic steps
resource_heavy_for_domain	65	High — token overuse relative to task type
model_downgrade_available	41	Low — smaller model likely sufficient
poor_agentic_control	31	Medium — tighten instructions, reduce tool breadth
overkill_for_agentic	21	— workflow may not need AI at all

Top optimization lever: Moving data-gathering from agentic turns to deterministic pre-steps (e.g., writing data to /tmp/gh-aw/agent/ in frontmatter) could eliminate up to 50% of token consumption across 68 runs.

Workflows with Errors

Workflow	Errors	Runs	Error Rate
Smoke Update Cross-Repo PR	3	2	150%
Smoke Create Cross-Repo PR	2	2	100%
Daily Syntax Error Quality Check	1	1	100%
Documentation Noob Tester	1	1	100%
Smoke Multi PR	1	1	100%
Release	1	3	33%
Smoke Service Ports	1	3	33%
The Great Escapi	1	1	100%
Daily News	1	1	100%
Stale Repository Identifier	1	1	100%
Smoke Copilot	1	11	9%
Super Linter Report	1	1	100%
CI Optimization Coach	1	1	100%

The Smoke suite shows persistent failures in cross-repo PR workflows — these should be investigated as infrastructure issues rather than token optimization.

Per-Workflow Full Statistics (All 65 Workflows)

#	Workflow	Total Tokens	Runs	Avg Tok/Run	Total Turns	Avg Turns/Run	Errors
1	Daily Syntax Error Quality Check	11,378,378	1	11,378,378	168	168.0	1
2	Contribution Check	6,734,002	6	1,122,334	204	34.0	—
3	Daily Community Attribution Updater	4,825,564	1	4,825,564	97	97.0	—
4	Daily Testify Uber Super Expert	4,492,441	1	4,492,441	62	62.0	—
5	Auto-Triage Issues	4,412,438	11	401,131	104	9.5	—
6	Documentation Noob Tester	3,742,876	1	3,742,876	69	69.0	1
7	Dead Code Removal Agent	3,670,585	1	3,670,585	52	52.0	—
8	Issue Monster	3,460,298	20	173,015	100	5.0	—
9	Daily Compiler Quality Check	3,277,331	1	3,277,331	46	46.0	—
10	Code Simplifier	2,821,647	1	2,821,647	45	45.0	—
11	Copilot CLI Deep Research Agent	2,666,932	1	2,666,932	36	36.0	—
12	Smoke Multi PR	2,663,743	1	2,663,743	59	59.0	1
13	jsweep - JavaScript Unbloater	2,611,045	1	2,611,045	35	35.0	—
14	Release	2,364,393	3	788,131	68	22.7	1
15	Workflow Health Manager - Meta-Orchestrator	2,107,676	1	2,107,676	29	29.0	—
16	Glossary Maintainer	1,977,266	1	1,977,266	28	28.0	—
17	Slide Deck Maintainer	1,906,240	1	1,906,240	31	31.0	—
18	PR Triage Agent	1,855,853	4	463,963	42	10.5	—
19	Daily Regulatory Report Generator	1,823,604	1	1,823,604	28	28.0	—
20	CI Cleaner	1,800,846	1	1,800,846	36	36.0	—
21	Plan Command	1,592,450	2	796,225	38	19.0	—
22	Dependabot Dependency Checker	1,435,143	1	1,435,143	22	22.0	—
23	Agent Container Smoke Test	1,419,341	10	141,934	39	3.9	—
24	Daily Semgrep Scan	1,311,393	1	1,311,393	25	25.0	—
25	Delight	1,225,216	1	1,225,216	21	21.0	—
26	Copilot PR Conversation NLP Analysis	1,165,538	1	1,165,538	23	23.0	—
27	The Daily Repository Chronicle	1,103,690	1	1,103,690	19	19.0	—
28	Daily MCP Tool Concurrency Analysis	1,049,376	1	1,049,376	20	20.0	—
29	GitHub Remote MCP Authentication Test	1,044,194	1	1,044,194	29	29.0	—
30	GPL Dependency Cleaner (gpclean)	938,414	1	938,414	21	21.0	—
31	CLI Consistency Checker	899,506	1	899,506	15	15.0	—
32	Dev	889,882	1	889,882	25	25.0	—
33	Daily Malicious Code Scan Agent	885,560	1	885,560	18	18.0	—
34	Terminal Stylist	876,457	1	876,457	16	16.0	—
35	Daily Copilot PR Merged Report	834,246	1	834,246	12	12.0	—
36	Daily Secrets Analysis Agent	779,712	1	779,712	17	17.0	—
37	Daily CLI Performance Agent	728,026	1	728,026	15	15.0	—
38	Draft PR Cleanup	665,354	1	665,354	17	17.0	—
39	Architecture Diagram Generator	665,216	1	665,216	14	14.0	—
40	Repository Quality Improvement Agent	609,756	1	609,756	12	12.0	—
41	Smoke Update Cross-Repo PR	555,987	2	277,994	19	9.5	3
42	Daily Project Performance Summary Generator	530,842	1	530,842	11	11.0	—
43	Daily Safe Output Integrator	480,560	1	480,560	9	9.0	—
44	Smoke Service Ports	458,444	3	152,815	13	4.3	1
45	Copilot PR Prompt Pattern Analysis	445,586	1	445,586	10	10.0	—
46	Daily Workflow Updater	377,836	1	377,836	10	10.0	—
47	Sub-Issue Closer	371,649	1	371,649	10	10.0	—
48	Smoke Create Cross-Repo PR	300,972	2	150,486	12	6.0	2
49	Issue Triage Agent	273,450	1	273,450	8	8.0	—
50	Constraint Solving — Problem of the Day	150,000	1	150,000	4	4.0	—
51	The Great Escapi	77,743	1	77,743	2	2.0	1
52–65	(14 workflows with 0 tokens — infra/scheduled/not-yet-run)	0	varies	0	0	—	—

---

🔍 Methodology & Data Quality

Methodology Notes

• Data source: Pre-downloaded workflow run logs from github/gh-aw (last 30 days)
• Engine filter: Copilot agent only (agent: copilot)
• Reporting date: 2026-04-01 UTC
• Historical data: 7 snapshots tracked since 2026-01-15
• Cost data: Not available in this dataset (all entries show $0.00)
• Memory storage: /tmp/gh-aw/repo-memory/default/memory/token-metrics/history.jsonl
• Analysis tool: Python (pandas, matplotlib, seaborn)
• 14 workflows had zero token usage — these appear to be infrastructure workflows (Smoke Copilot, Q, metrics collectors) that either don't consume inference tokens or failed before agent activation

References: §23846022060

AI generated by Daily Copilot Token Consumption Report · history

• expires on Apr 4, 2026, 11:41 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 #24071.