[api-consumption] 📊 GitHub API Consumption Report — 2026-06-17

[github-actions[bot]]

📊 GitHub API Consumption Report

Report Date: 2026-06-17 · Repository: github/gh-aw · Run: #27687758429

---

Today at a Glance

Metric	Value
🤖 Total Runs	391 (302 ✅ / 68 ❌)
🎯 Success Rate	77.2%
🔗 GitHub REST API Calls	574,243 (core quota consumed — includes reads, writes, and all GitHub API operations)
📝 Safe-Output Writes	210 (issues + PRs + comments + discussions created by safe-output tools)
⏱ Avg Duration	615s (p95: 1,394s)

Note on scale: the 24h window spans 2026-06-16 12:27Z → 2026-06-17 12:27Z (the same rolling-window convention used by previous entries). Today's REST API total is the highest in the 90-day window — see the trend narrative below for why the prior two days read anomalously low.

---

🔗 GitHub API Calls Trend (90 days)

The daily baseline normally sits around 40k–55k REST calls, with an earlier peak of 207,880 on 2026-06-09. Today's reading of 574,243 is a sharp spike — roughly 2.8× the previous high. Crucially, the two immediately preceding days (06-15: 2,411 and 06-16: 3,216) are anomalous lows caused by incomplete log capture in those runs, so part of today's jump is a return to (and beyond) the normal range rather than a pure surge. The 7-day rolling average has turned upward as a result.

---

🔗 GitHub API Calls by Workflow Trend (30 days)

The heaviest sustained consumers over the last 30 days are the PR-review and CI smoke workflows — Smoke CI, PR Code Quality Reviewer, Design Decision Gate, and Matt Pocock Skills Reviewer repeatedly top the per-workflow series. Today these same workflows surged together, which is the main driver behind the aggregate spike. Most other workflows stay flat in the low-thousands, confirming that API pressure is concentrated in a small, predictable set of agents.

---

🔗 GitHub REST API Calls Heatmap (90 days)

Consumption is spread fairly evenly across weekdays, with mid-week (Tue–Thu) running slightly hotter as PR and CI activity peaks. The current ISO week stands out brightly because of today's elevated total, while the late-second-half of the prior week shows the dim cells corresponding to the under-collected 06-15/06-16 entries. No consistent weekend lull is visible — scheduled daily workflows keep the floor populated every day.

---

🍩 Top API Burners (24h)

The top 3 workflows (Smoke CI, Matt Pocock Skills Reviewer, Design Decision Gate) alone account for 35% of all REST consumption, and the top 8 named workflows make up 61%. The remaining 135 workflows together contribute the other 39%. This is a moderate concentration: tuning just the handful of heavy reviewers would meaningfully lower the repository-wide footprint.

---

🔗 GitHub REST API Consumption by Workflow (last 24h)

Smoke CI leads at 84,545 calls across 35 runs, followed by the review agents in the 25k–58k range. Notably, 15 individual runs each consumed more than 10,000 core calls and 28 more landed in the 5k–10k band; the single heaviest run (Matt Pocock Skills Reviewer, 11,798) came within ~21% of the 15,000/hr rate-limit ceiling. These near-limit runs are the prime optimisation targets — caching collaborator/PR lookups and batching GraphQL where possible would pull them well clear of the limit.

---

Top 10 Workflows by REST API Consumption (last 24h)

Workflow	REST API Calls	Runs	Avg Duration
Smoke CI	84,545	35	288s
Matt Pocock Skills Reviewer	58,192	21	592s
Design Decision Gate 🏗️	57,795	21	394s
PR Code Quality Reviewer	42,052	18	709s
Test Quality Sentinel	33,683	21	618s
PR Description Updater	26,279	25	537s
PR Sous Chef	25,405	17	504s
Smoke Copilot - AOAI (Entra)	23,939	6	753s
Smoke Pi	22,432	7	285s
Agent Container Smoke Test	21,202	7	350s

---

Trending Indicators

• 7-day API trend: ↑ +25.7% vs. previous 7 days (inflated by today's spike and the two under-collected prior days)
• 30-day API trend: → insufficient data (history holds 30 days; a full 30-vs-30 comparison needs 60)
• GitHub REST API call rate: ~98,206 calls/day over the last 7 days (hourly limit: 15,000) — heavily weighted by today's outlier

---

📦 Cache Memory Status

• Location: /tmp/gh-aw/cache-memory/trending/api-consumption/history.jsonl
• Cache restored from previous run: yes (29 entries)
• Collection mode: incremental (-1d) — history was already dense (29 days), so a full 90-day backfill was unnecessary; collecting the last 24h brings the merged history to the 30-day minimum needed for trend math
• Logs start_date used: -1d
• Data points stored: 30
• Earliest entry: 2026-05-18
• Retention policy: 90 days

⚠️ Data-quality notes

• The agenticworkflows logs MCP call repeatedly timed out at the 120s bridge limit, but downloaded run directories incrementally across passes. Final aggregation used 391 complete runs (those with both run_summary.json and aw_info.json) spanning the exact rolling 24h window; partially-downloaded directories were excluded.
• github_api_calls is the sum of per-run github_rate_limit_usage.core_consumed (response-header delta). Concurrent runs sharing a rate-limit bucket can inflate individual deltas, so treat the absolute total as an upper-bound indicator of pressure rather than an exact call count.
• The 06-15 and 06-16 history entries (2,411 and 3,216) are believed to be under-collected by prior runs, which exaggerates today's apparent week-over-week jump.

---

Automatically generated by the api-consumption-report workflow.

Generated by 📊 GitHub API Consumption Report Agent · ◷

• expires on Jun 20, 2026, 4:46 AM UTC-08:00