[lockfile-stats] Lockfile Statistics Audit — 2026-10-26

[github-actions[bot]]

Overview

Analyzed 349 compiled .github/workflows/*.lock.yml files (0 skipped/malformed) via single-script compact JSON analysis. Total compiled size: 62.74 MB (avg 179.8 KB, min 98.1 KB, max 339.5 KB per lockfile).

Key metrics

Metric	Value
Lockfiles	349 (+1 vs 2026-10-25)
Total bytes	62,736,235 (+313,398)
Avg jobs/workflow	9.07 (max 25, min 6)
Avg steps/workflow	157.2 (max 343, total 54,866)
Avg script steps/workflow	108.9 (max 196, total 38,021)
workflow_dispatch enabled	339 / 349 (97.1%)
schedule enabled	253 / 349 (72.5%)

Trigger analysis

Top combinations: schedule+workflow_dispatch (216), workflow_dispatch only (73), pull_request+schedule+workflow_dispatch (33), pull_request+workflow_dispatch (9). Most common cron: 0 0 */2 * * (every 2 days) on 61 workflows; the remaining ~190 cron expressions are essentially unique.

Safe outputs analysis

Standard safe-output scaffolding (noop, report_incomplete, report_missing_data, report_missing_tool) appears in 342/349 workflows (98%). Most common functional safe outputs: create_issue (185), create_discussion (92), add_comment (74), create_pull_request (66).

Discussion category detection: 93 workflows configure create_discussion; 92 resolved a category (98.9%), 0 required fallback regex parsing. Category breakdown: audits (79), announcements (5), artifacts (2), dev (2), research (2), general (1), daily-news (1). One workflow's create_discussion config did not resolve a category string — within the 10% detection tolerance, so this is a minor gap, not a detection failure (safe_outputs_config_missing = 0, yaml_available = true).

Permission patterns (agent job only)

contents: read is universal (349/349); the agent job itself never holds write for any tracked scope (contents/issues/pull-requests/discussions/actions/packages/deployments/checks/statuses/security-events/pages all show 0 write in the agent job). Only id-token: write appears directly on 4 agent jobs.

By contrast, the union across all jobs in each workflow shows pervasive write capability: id-token write in all 349 workflows, issues write in 346, contents write in 232, pull-requests write in 153, discussions write in 97 — confirming writes are consistently delegated to separate safe-output jobs rather than the agent job, matching the intended privilege-separation design. agent_job_permissions_unknown = 0.

Engine distribution (from gh-aw-metadata agent_id)

Engine	Count	Share
copilot	145	41.5%
codex	77	22.1%
claude	63	18.1%
pi	38	10.9%
aider / goose	4 each	1.1%
agy / opencode	3 each	0.9%
crush / cursor / deepseek-harness / gemini / kiro / pydantic-ai	2 each	0.6%

engine_unknown: 0 (all 349 lockfiles resolved an engine id, no fallback needed).

Tool & MCP patterns

Top MCP servers (parsed from gh-aw-manifest, 0 legacy fallback scrapes needed): safeoutputs (349, universal), github (208), agenticworkflows (44), work-queue (37), mcpscripts (33), serena (24), tavily / sentry (5 each), agy-native / grafana (3 each).

Interesting findings

1. Mass timeout policy shift: job timeout-minutes buckets moved sharply between the last two runs — the 45-minute bucket dropped from 345 → 72 workflows while the 15-minute bucket rose from 1 → 275. That's a ~273-job swing, consistent with a deliberate batch tightening of a default timeout from 45 to 15 minutes rather than organic per-workflow drift.
2. Agent jobs are structurally incapable of direct writes — every write-capable scope shows 0 in the agent-job permission counts, while the same scopes show heavy write counts in the union-across-jobs view (e.g. contents write: 0 direct vs 232 via union). This is strong evidence the safe-outputs privilege-separation pattern is applied consistently repo-wide, not just in a few workflows.
3. copilot leads at 41.5% of workflows, but the engine tail is wide: 14 distinct engine ids appear, including niche ones like pydantic-ai, agy, and deepseek-harness at 2-3 workflows each.
4. Despite 253 workflows using schedule, cron timing is highly fragmented — only one cron expression (0 0 */2 * *, 61 workflows) is shared by more than 3 workflows; the rest are essentially unique, suggesting schedules were hand-set per workflow rather than inherited from a shared template.
5. create_issue (185) is used in more than half of all workflows — roughly double the next most common functional safe output (create_discussion, 92) — making issue creation the dominant non-scaffolding safe output across the fleet.

Historical trends (vs 2026-10-25)

Metric	2026-10-25	2026-10-26	Δ
Lockfiles	348	349	+1
Total bytes	62,422,837	62,736,235	+313,398
workflow_dispatch	338	339	+1
create_issue	184	185	+1
codex engine	76	77	+1
45-min timeout jobs	345	72	-273
15-min timeout jobs	1	275	+274

The timeout-bucket swing is the standout change this run; all other deltas are consistent with a single new workflow being added, with no other structural shifts.

Recommendations

• Confirm the 45→15 minute timeout change was intentional; watch for new timeout-related job failures in workflows swept into the shorter bucket.
• Investigate the one create_discussion workflow whose category did not resolve, to bring detection back to 100%.
• Continue auditing that no agent job gains direct write permissions as new workflows are added — the current 0% direct-write rate on agent jobs is a security invariant worth protecting going forward.

Methodology note: single-script compact JSON analysis — one Python script parsed all lockfiles via yaml.safe_load in a single run, with engine/permission/safe-output detection driven by gh-aw-metadata, gh-aw-manifest, and GH_AW_SAFE_OUTPUTS_CONFIG rather than frontmatter assumptions. yaml_available: true; 0 skipped lockfiles; 0 fallback parses needed.

References:

• §40395199133

Generated by 📊 Lockfile Statistics Analysis Agent · claude · agent · ~55 AIC · ⌖ ~1 AIC · ⊞ 4.3K · ◷

• expires on Oct 27, 2026, 12:31 PM UTC-08:00

[github-actions[bot]]

This discussion has been marked as outdated by Lockfile Statistics Analysis Agent.

A newer discussion is available at Discussion #72509.