[lockfile-stats] Lockfile Statistics Report — 2026-09-10 (299 workflows analyzed)

[github-actions[bot]]

Lockfile Statistics — 2026-09-10

Analysis of all .github/workflows/*.lock.yml in github/gh-aw, generated by run §34525720125.

299 lockfiles, 0 skipped, 45,873,733 bytes total (avg 153.4 KB, min 90.0 KB, max 239.6 KB).

Key findings

1. Split-permission architecture is universal. Every agent job caps out at contents: read plus a handful of scoped reads (issues 260/299, pull-requests 255/299, actions 114/299, discussions 52/299) — no agent job holds a write scope. Yet the union of all jobs in a workflow shows a write scope in all 299/299 workflows. This confirms every workflow uses a dedicated, more-privileged "apply safe-outputs" job rather than granting the agent job write access directly.
2. Baseline safe-output scaffolding is near-universal, real writes are selective. noop/missing_data/missing_tool/report_incomplete/create_report_incomplete_issue each appear in 293/299 workflows (98%), but actual write-producing types are far rarer: create_issue 146 (48.8%), create_discussion 92 (30.8%), add_comment 77 (25.8%), create_pull_request 64 (21.4%).
3. Discussion-category detection was 100% clean — 92/92 create_discussion workflows resolved a category, 0 via fallback, 0 unresolved. Of those, 79 (85.9%) post to audits, with the rest spread across announcements (5), artifacts (2), dev (2), research (2), general (1), daily-news (1).
4. Engine adoption: copilot leads at 123/299 (41.1%), then codex 74 (24.7%), claude 56 (18.7%), pi 31 (10.4%), with a long tail of 9 other engines (aider, opencode, goose, crush, cursor, deepseek-harness, gemini, kiro, pydantic-ai) at ≤3 workflows each. engine_unknown = 0 — every lockfile resolved an engine from gh-aw-metadata.
5. No change since 2026-09-09. Every metric in this report is byte-identical to yesterday's snapshot — the lockfile fleet has been stable for at least 24h.

Trigger analysis

Combination	Count	Share
schedule + workflow_dispatch	209	69.9%
workflow_dispatch only	38	12.7%
pull_request + schedule + workflow_dispatch	29	9.7%
pull_request + workflow_dispatch	9	3.0%
pull_request only	3	1.0%
other combinations (10 distinct)	11	3.7%

workflow_dispatch appears in 291/299 (97.3%); schedule in 242/299 (81.0%). The single most common cron is 0 0 */2 * * (every 2 days at 00:00 UTC) at 42 workflows — a potential scheduling thundering-herd (see recommendations).

Full trigger and cron breakdown

Triggers: workflow_dispatch 291, schedule 242, pull_request 43, issues 5, issue_comment 3, workflow_call 3, deployment_status 1, workflow_run 1, discussion 1, discussion_comment 1, push 1, pull_request_review_comment 1.

Notable crons (>1 occurrence): 0 0 */2 * *×42, 7 5 * * *×3, then eight distinct crons at ×2, and 20 more at ×1.

Safe outputs analysis

Type	Count	Type	Count
create_issue	146	create_pull_request_review_comment	16
create_discussion	92	submit_pull_request_review	16
add_comment	77	upload_artifact	14
create_pull_request	64	create_check_run	14
push_repo_memory	36	update_issue	11
add_labels	32	push_to_pull_request_branch	11
upload_asset	26	max_bot_mentions	8
mentions	23	close_issue / remove_labels	7 each

safe_outputs_config_missing = 0 — configuration was recovered for every lockfile. Long tail (≤6 workflows each) includes emerging integrations: jira_create_issue, linear_create_issue, create_project_status_update, send-slack-message/post_slack_message variants.

Structural characteristics

Metric	Avg	Max
Jobs / workflow	7.35	14
Steps / workflow	144.15	205
Script (run:) steps / workflow	98.72	136

Permission patterns (agent job vs. union of all jobs)

Scope	Agent job: read / write / none	Union of jobs: read / write / none
contents	299 / 0 / 0	93 / 206 / 0
issues	260 / 0 / 39	0 / 299 / 0
pull-requests	255 / 0 / 44	116 / 146 / 37
actions	114 / 0 / 185	190 / 109 / 0
discussions	52 / 0 / 247	10 / 96 / 193
security-events	14 / 0 / 285	12 / 3 / 284
id-token	0 / 2 / 297	0 / 3 / 296

permissions_unknown = 0. All other scopes (checks, deployments, packages, pages, repository-projects, statuses, attestations) are none for ≥298/299 agent jobs. Note issues: write in 299/299 union — the safe-outputs apply job universally needs it (e.g. for label/comment side effects), independent of whether create_issue itself is configured.

Engine distribution

copilot 123 · codex 74 · claude 56 · pi 31 · aider 3 · opencode 3 · goose 3 · crush 1 · cursor 1 · deepseek-harness 1 · gemini 1 · kiro 1 · pydantic-ai 1

Top models: openai/gpt-5.3-codex (37), copilot/gpt-5.3-codex (33), copilot/auto (19), openai/gpt-5.4 (11), copilot/claude-sonnet-4.5 (9).

Tool & MCP patterns

safeoutputs MCP server is present in all 299 workflows; github in 175 (58.5%); agenticworkflows in 44; serena in 25; mcpscripts in 12; smaller usage of tavily, sentry, ast-grep, datadog, deepwiki, microsoftdocs, grafana, kreuzberg, mempalace, graft, agentdb, ruflo (≤5 each). mcp_fallback_used_count = 0 — every server list came from the gh-aw-manifest JSON, no regex fallback needed.

Historical trends

Compared against /tmp/gh-aw/cache-memory/history/2026-09-09.json: no deltas — lockfile count, byte totals, trigger/safe-output/engine/permission distributions are all identical to yesterday.

Recommendations

• Stagger the 42 workflows sharing the 0 0 */2 * * cron across the day to avoid a midnight-UTC scheduling spike.
• The agent job → read-only / apply job → write permission split is consistently applied fleet-wide; no gaps found worth flagging.
• Watch the small but growing set of third-party integrations (Jira, Linear, Slack variants) in the safe-output long tail — worth a shared template if adoption continues.

Methodology: single-script compact JSON analysis (lockfile_stats_v4.py, cached at /tmp/gh-aw/cache-memory/scripts/), reading gh-aw-metadata/gh-aw-manifest header comments and the GH_AW_SAFE_OUTPUTS_CONFIG env JSON per lockfile — never the always-empty top-level permissions: {}.

References:

• §34525720125

Warning

Firewall blocked 1 domain

The following domain was blocked by the firewall during workflow execution:

• api.anthropic.com

To allow these domains, add them to the network.allowed list in your workflow frontmatter:

network:
  allowed:
    - defaults
    - "api.anthropic.com"

See Network Configuration for more information.

Generated by 📊 Lockfile Statistics Analysis Agent · claude · agent · 80 AIC · ⌖ 9.45 AIC · ⊞ 8.5K · ◷

• expires on Sep 11, 2026, 12:28 PM UTC-08:00