[lockfile-stats] Lockfile Statistics Report — 2026-06-03 #36733
Closed
Replies: 1 comment
-
|
This discussion has been marked as outdated by Lockfile Statistics Analysis Agent. A newer discussion is available at Discussion #36994. |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Executive Summary
Analysis of 239 compiled
.lock.ymlworkflows in.github/workflowsas of 2026-06-03. All files parsed successfully (0 skipped).test-workflow.lock.yml— 65.5 KBsmoke-claude.lock.yml— 184.9 KBThe corpus grew by one workflow since 2026-06-02 — a scheduled,
workflow_dispatch-enabled Copilot-engine workflow contributing ~7 jobs, ~100 steps, and ~50 inline scripts.File Size Distribution
Lockfiles are large and tightly clustered (compiled output, not source). No file is below 50 KB.
Top 10 largest lockfiles
Trigger Analysis
workflow_dispatchis near-universal (231/239, 96.7%) andscheduledrives the majority (161/239, 67.4%) — consistent with a fleet of mostly autonomous, cron-driven agents that also retain manual-run capability.Top trigger combinations
The dominant pattern (
schedule + workflow_dispatch, 65.7% of all workflows) confirms the "scheduled agent with manual override" archetype.Schedule cadence notes
30 distinct cron expressions in use. Most are once-daily at randomized minutes/hours (jitter to avoid API thundering-herd). A few run more frequently:
23 * * * *(hourly), several*/6/*/4-hour cadences, and weekday-only business-hours schedules (* * 1-5).Structural Characteristics
Compiled workflows are highly uniform in shape.
Totals: 1,911 jobs, 25,285 steps, 12,131 inline scripts across the corpus. The narrow job range (5–12) reflects the shared compiler scaffold; variation comes mostly from engine and safe-output plumbing.
Timeout Patterns
The vast majority of jobs cap at 30 minutes or less — sensible bounding for agentic runs.
Engine Distribution
Copilot is the default engine for ~66% of workflows; Claude is the clear second at ~26%.
Tool & MCP Patterns
The GitHub MCP server dominates tool wiring with 6,552 tool references — roughly 27 GitHub tools enabled per workflow on average. Many GitHub read tools appear in exactly 126 workflows, indicating a shared read-only GitHub toolset template.
Note on safe-output & permission extraction
The cached v1 analyzer reports empty
safe_output_types,discussion_categories, and top-levelpermissionsmaps (all 239 workflows show a{}top-level permissions kind). Because the compiled lockfiles embed these as job-level GitHub Actions constructs rather than the source-levelsafe-outputs:keys, the current regex-based pass does not resolve them. This is a known limitation of the v1 schema and a candidate for a v2 bump — flagged under Recommendations rather than reported as zero usage.Interesting Findings
smoke-*engine matrix workflows, the natural consequence of exercising multiple engine paths in one lockfile.workflow_dispatch, so even fully autonomous scheduled agents remain hand-runnable.Historical Trends (2026-06-02 → 2026-06-03)
The single net-new workflow is scheduled + dispatch-enabled and Copilot-powered, matching the corpus's dominant archetype. 15 days of history are retained (2026-05-20 onward).
Recommendations
lockfile_stats_v2.pyto resolve safe-output types, discussion categories, and effective permissions from the compiled job structure (current v1 reports these as empty due to regex limits).Methodology
Single-script compact JSON analysis: one cached Python analyzer (
lockfile_stats_v1.py) parsed all 239 lockfiles in a single pass, emitting a compact summary JSON (4.7 KB); all insights and deltas were derived from that summary and the prior day's cached summary. No per-file content was re-read during reporting.References: §26913585542
Beta Was this translation helpful? Give feedback.
All reactions