Artifacts Usage Report — github/gh-aw

[github-actions[bot]]

Overview

This report summarizes GitHub Actions artifact usage across the github/gh-aw repository. The repository has 267 registered workflows and over 481,000 total workflow runs. Analysis focused on active workflows with recent runs and measured per-run artifact profiles.

Key finding: Artifact generation is concentrated in the Smoke test suite (Claude, Codex, Copilot, OpenCode) which runs on every pull request. The agent artifact (workflow session logs/context) and detection artifact (threat/content analysis) are the primary persistent storage consumers. Most workflows produce short-lived activation artifacts (1-day TTL) that do not accumulate.

---

Artifact Summary Table

Workflow Name	Runs (Total)	Artifacts/Run	Avg Size/Run (total)	Persistent/Run (90d)	Latest Run	Status
Smoke Copilot	3,102	7	35.5 MB	0.49 MB	2026-04-26	✅ Active
Smoke Codex	2,966	4	1.68 MB	1.65 MB	2026-04-26	✅ Active
Smoke Claude	2,980	7	0.46 MB	0.43 MB	2026-04-26	✅ Active
Contribution Check	(per-PR)	4	0.56 MB	0.54 MB	2026-04-26	✅ Active
Daily News	191	3	1.33 MB	0.08 MB	2026-04-24	✅ Active
Daily CLI Performance Agent	~191	5	0.17 MB	0.14 MB	2026-04-26	✅ Active
Dictation Prompt Generator	(per-event)	4	0.17 MB	0.15 MB	2026-04-26	✅ Active
Smoke OpenCode	189	2	0.09 MB	0.07 MB	2026-04-26	✅ Active
Agentic Maintenance	(scheduled)	1	< 0.01 MB	< 0.01 MB	2026-04-26	✅ Active
CI	24,111	0	—	—	2026-04-26	✅ Active
Scout	34,587	0	—	—	2026-04-26	✅ Active
Doc Build - Deploy	8,070	0	—	—	2026-04-26	✅ Active
Archie, Q, /cloclo, AI Moderator	high	0	—	—	2026-04-26	✅ Active

Note: Per-run sizes are sampled from single recent runs. Actual sizes vary by workflow content.

---

Artifact Type Breakdown

Standard artifact types produced by agentic (lock.yml) workflows

Artifact Name	Typical Size	TTL	Description
activation	22–31 KB	1 day	Activation job context; short-lived
agent	74 KB – 1.5 MB	90 days	Agent session logs and workspace (primary storage)
detection	3 – 185 KB	90 days	Threat detection / content moderation results
safe-outputs-items	< 1 KB	90 days	Safe output metadata
repo-memory-default	2 KB – 1.28 MB	1 day	Cache-memory for cross-run persistence
gh-aw (binary)	~26.6 MB	1 day	Built binary (Smoke Copilot only)
safe-outputs-upload-artifacts	~10 MB	1 day	Safe output upload payload (Smoke Copilot only)
dockerbuild	~27 KB	90 days	Docker build metadata (Smoke Claude, Copilot)
code-scanning-sarif	~557 B	1 day	Security scan results (Smoke Claude)
apm	~3.4 KB	1 day	APM telemetry (Smoke Claude)
secret-validation-report	~1.2 KB	30 days	Secret rotation validation (Agentic Maintenance)

---

Top Storage Consumers

Workflows generating the most persistent artifact storage (90-day window):

1. 🔴 Smoke Codex — 1.65 MB/run × ~370 runs/month ≈ 610 MB/month persistent

   • Dominated by large agent artifacts (~1.5 MB each, containing full Codex session logs)
   • detection artifacts unusually large (~185 KB vs. ~3 KB for other smoke tests)

2. 🟠 Smoke Copilot — 0.49 MB/run × ~388 runs/month ≈ 190 MB/month persistent

   • Binary artifact (gh-aw, ~26.6 MB) and safe-output upload (~10 MB) expire in 1 day, limiting accumulation
   • Persistent portion is modest: agent (~485 KB) + dockerbuild (~27 KB)

3. 🟡 Smoke Claude — 0.43 MB/run × ~372 runs/month ≈ 160 MB/month persistent

   • detection artifacts notably larger (~27 KB) vs. other smoke tests (~3 KB)

4. 🟢 Contribution Check — 0.54 MB/run (per-PR frequency) — variable total

5. 🟢 Daily News — ~81 KB/run × ~24 runs/month ≈ 1.9 MB/month persistent

---

Insights & Recommendations

💡 Optimization Opportunities

1. Smoke Codex agent artifacts are 3–4× larger than other smoke tests (~1.5 MB vs. ~400–500 KB). Investigate whether Codex session output can be truncated or compressed before upload. This workflow accounts for the single largest persistent storage category.

2. Smoke Copilot uploads gh-aw binary (~26.6 MB) and safe-outputs-upload (~10 MB) per run — these expire after 1 day which is appropriate. No action needed, but worth monitoring if TTL policies change.

3. repo-memory-default artifacts in Daily News (~1.28 MB) expire in 1 day — this is correct behavior for cross-run cache-memory. Functioning as designed.

4. Activation artifacts (1-day TTL) are created by every agentic workflow — these are intentionally short-lived and not a storage concern.

5. High-volume workflows (CI: 24K runs, Scout: 34K runs, Doc Build: 8K runs) generate zero artifacts — excellent. These are well-optimized.

6. 267 total workflows, majority generate no artifacts — the repo follows a healthy "artifacts only when needed" pattern.

📊 Estimated 90-Day Artifact Accumulation

Based on run frequency and 90-day retention:

Workflow	Runs in 90d (est.)	Persistent Size (est.)
Smoke Codex	~1,110	~1.83 GB
Smoke Claude	~1,116	~480 MB
Smoke Copilot	~1,164	~570 MB
Smoke OpenCode	~87	~6.4 MB
Daily News	~90	~7.3 MB
Daily CLI Perf	~90	~12.6 MB
Contribution Check	variable	~variable
All others	low	< 50 MB
Total estimate		~2.9–3.5 GB

These are rough estimates; actual storage depends on GitHub's artifact retention enforcement and exact run frequency.

---

Recommendations

1. Investigate Smoke Codex agent artifact size — at 1.5 MB/run, this is the top storage concern. Consider compressing logs or filtering verbose output before artifact upload.

2. Review detection artifact size for Smoke Codex — at 185 KB vs. 3–27 KB for other smoke tests, this may indicate a configuration or output verbosity difference.

3. Monitor total storage usage via GitHub's billing/storage UI to validate these estimates against actual usage.

4. No action needed for workflows without artifacts (CI, Scout, Doc Build, etc.) — these are correctly configured.

References:

• §24949827772 — Dictation Prompt Generator (sample)
• §24945190957 — Smoke Claude (sample)
• §24945190968 — Smoke Codex (sample)

Generated by Artifacts Summary · ● 1.4M · ◷

• expires on Apr 27, 2026, 6:24 AM UTC

[github-actions[bot]]

This discussion was automatically closed because it expired on 2026-04-27T06:24:49.503Z.

Closed by Workflow