[lockfile-stats] Lockfile Statistics — 2026-08-28 (297 workflows)

[github-actions[bot]]

📊 Lockfile Statistics — 2026-08-28

Analysis of 297 compiled .github/workflows/*.lock.yml files in github/gh-aw (0 malformed/skipped). Compared against the prior snapshot from 2026-08-27 (295 lockfiles).

Run: §33214946647

Key metrics

Metric	Value	Δ vs 2026-08-27
Lockfiles	297	+2
Total size	45,031,678 bytes (43.0 MB)	+77,225 bytes
Avg size	151,621.8 bytes	-766.2
Min / Max size	90,270 / 239,827 bytes	+127 / +40
workflow_dispatch present	289	+2
schedule present	240	+1
Workflows granting any job write permission	297 (100%)	+2

File size distribution

• Range: 90,270 – 239,827 bytes; mean 151,621.8 bytes.
• Sizes are tightly clustered relative to the range — no outliers beyond the max/min shift of a couple hundred bytes since yesterday, consistent with normal content-only edits rather than structural changes.

Trigger analysis

Top trigger combinations:

Combination	Count
schedule+workflow_dispatch	207
workflow_dispatch only	38
pull_request+schedule+workflow_dispatch	29
pull_request+workflow_dispatch	9
pull_request only	3

Cron cadence: the dominant cron is 0 0 */2 * * (every 2 days) at 42 workflows; the rest of the 30 distinct cron strings are singletons or small clusters (2-3 workflows), indicating most schedules are individually tuned rather than copy-pasted.

A new workflow_call+workflow_dispatch combination appeared (1 workflow) that wasn't present yesterday, alongside the workflow_call trigger count rising from 2 → 3.

Safe outputs analysis

92 workflows configure create_discussion (unchanged from yesterday); all 92 resolved to a discussion category (0 unresolved, 0 fallback-parsed) — detection is exhaustive, not partial.

Discussion categories:

Category	Count
audits	79
announcements	5
artifacts	2
dev	2
research	2
general	1
daily-news	1

Most common safe-output types (beyond the universal missing_data/missing_tool/noop/report_incomplete/create_report_incomplete_issue, each ~291): create_issue (145, +2), create_discussion (92), add_comment (77), create_pull_request (63), push_repo_memory (34).

safe_outputs_config_missing: 0 — every lockfile had a parseable GH_AW_SAFE_OUTPUTS_CONFIG.

Structural characteristics

	Avg	Max
Jobs/workflow	7.33	14
Steps/workflow	143.89	202
Script-bearing steps/workflow	99.15	133

Step and script averages both ticked down slightly (144.09→143.89, 99.26→99.15) despite 2 new workflows joining — the new workflows are leaner than the existing average.

Permission patterns (agent job)

permissions_unknown: 0 — every workflow's agent job permissions resolved cleanly.

Scope	read	write	none
contents	297	0	0
issues	259	0	38
pull-requests	254	0	43
actions	112	0	185
discussions	52	0	245
security-events	14	0	283
id-token	0	2	295

The agent job itself never requests direct write scopes except id-token (2 workflows) — writes happen through the safe-outputs mechanism, not the agent job's own permissions. Across the union of all jobs per workflow, all 297 workflows (100%) grant at least one write scope somewhere in the job graph (typically issues: write — granted in all 297).

Engine distribution

Engine	Count	Δ
copilot	151	+3
claude	61	-1
codex	46	-1
pi	22	+1
crush	3	—
aider	3	—
opencode	3	—
goose	3	—
cursor	1	—
deepseek-harness	1	—
gemini	1	—
kiro	1	—
pydantic-ai	1	—

engine_unknown: 0 — every lockfile's agent_id resolved from gh-aw-metadata. Copilot remains the dominant engine at just over half of all workflows (151/297, 50.8%).

Tool & MCP patterns

MCP server	Workflows using it
safeoutputs	297 (universal)
github	121
agenticworkflows	44
serena	25
mcpscripts	12
tavily	5
sentry	3

mcp_fallback_used_count: 0 — all MCP usage was resolved from the gh-aw-manifest comment, no legacy scraping needed. The GitHub MCP server's most-used tools are broad read operations (get_commit, get_file_contents, list_commits, search_code, each ~119-121 workflows), consistent with workflows doing repo-wide investigation rather than narrow single-PR/issue lookups.

Interesting findings

1. 100% write coverage via safe outputs, 0% via direct agent permissions (except id-token). Every workflow's agent job is scoped to read-only (or none) for content/issue/PR scopes, and all mutating actions flow through the safeoutputs MCP server — a consistent security posture across the entire fleet.
2. Copilot's share is growing while Claude's and Codex's shrink slightly (+3 / -1 / -1 day-over-day) — a small but directionally consistent shift toward Copilot as the dominant engine.
3. Cron schedules are mostly bespoke. Aside from the 0 0 */2 ** cluster (42 workflows), the remaining ~200 scheduled workflows use one of 29 distinct, largely unique cron expressions — little copy-paste standardization on schedule timing.
4. Discussion category detection is fully exhaustive (92/92, 0 unresolved) — the GH_AW_SAFE_OUTPUTS_CONFIG JSON extraction path is reliable across the whole fleet, with zero reliance on regex fallback.
5. New workflows trend leaner. The two workflows added since yesterday pulled down both the step-count and script-count averages, despite raising max size only marginally (+40 bytes) — suggesting newer workflows are more compact than the historical norm.

Historical trends (vs 2026-08-27)

• Lockfiles: 295 → 297 (+2)
• Total bytes: 44,954,453 → 45,031,678 (+77,225)
• create_issue safe outputs: 143 → 145 (+2)
• schedule+workflow_dispatch combo: 206 → 207 (+1)
• copilot engine: 148 → 151 (+3); claude: 62 → 61 (-1); codex: 47 → 46 (-1); pi: 21 → 22 (+1)
• github MCP usage: 122 → 121 (-1)
• create_discussion/category detection: unchanged at 92/92

Recommendations

• Continue monitoring the copilot/claude/codex engine share shift — if it accelerates, worth understanding whether it's driven by new workflow authorship preference or migrations of existing workflows.
• No parser or detection issues found this run (engine_unknown, permissions_unknown, safe_outputs_config_missing, and discussion_category_unresolved all at 0) — the v4 analyzer schema remains stable and needs no changes.

Methodology note

single-script compact JSON analysis

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 · 50.6 AIC · ⌖ 7.34 AIC · ⊞ 8.5K · ◷

• expires on Aug 29, 2026, 2:08 PM UTC-08:00