
cron-schedule-planner
Builds an overview of GitHub Actions `schedule` cron expressions across multiple repositories and helps spread CPU-intensive jobs (Docker Compose, Testcontainers, Kind/k3s, matrix builds) into less contended slots. Use when the user asks 'when do my cron actions run?', 'do my scheduled builds collide?', 'when should the Testcontainers build run?', or wants a cron schedule overview/distribution report across repos. Scans via `gh` (no local checkouts required) or local checkouts, converts UTC to a local timezone, classifies job intensity, and writes a Markdown report with an hourly histogram, collisions and concrete cron redistribution suggestions.
Builds an overview of GitHub Actions `schedule` cron expressions across multiple repositories and helps spread CPU-intensive jobs (Docker Compose, Testcontainers, Kind/k3s, matrix builds) into less contended slots. Use when the user asks 'when do my cron actions run?', 'do my scheduled builds collide?', 'when should the Testcontainers build run?', or wants a cron schedule overview/distribution report across repos. Scans via `gh` (no local checkouts required) or local checkouts, converts UTC to a local timezone, classifies job intensity, and writes a Markdown report with an hourly histogram, collisions and concrete cron redistribution suggestions.
Cron Schedule Planner
Collects GitHub Actions on.schedule cron expressions across repositories,
classifies how resource-intensive each workflow is, converts the UTC cron times
to a local timezone, and produces a Markdown report with an hourly histogram,
collisions and concrete redistribution suggestions.
Read-only: the script never edits workflows. Suggestions are applied only after
the user confirms.
Instructions
Step 1: Decide the repository scope
- Default: all repositories of the authenticated
ghuser (no local checkouts
needed). --owner <owner>to scan a different owner.--repo <owner/repo>(repeatable) for an explicit list.--local <dir>to scan local checkouts instead of the GitHub API.
Step 2: Run the script
bash scripts/cron-schedule-planner.sh
Common variants:
# Specific owner, Swiss local time, report in target/
bash scripts/cron-schedule-planner.sh --owner dboeckli --tz Europe/Zurich
# Local checkouts (no GitHub API)
bash scripts/cron-schedule-planner.sh --local ~/projects/referenzen
# A few explicit repos
bash scripts/cron-schedule-planner.sh --repo dboeckli/ai-agent-skills --repo dboeckli/camel-first
Step 3: Present the results
The script prints the hourly histogram, collisions and suggestions, and writes
target/cron-schedule-overview.md. Summarise:
- how many repos / scheduled entries were found,
- the most crowded UTC hours (collisions),
- which high-intensity jobs sit in crowded hours.
Step 4: Apply suggestions only after confirmation
The redistribution suggestions are heuristic (they move high-intensity jobs out
of crowded hours into empty ones, keeping the minute and day fields). Present
them as a diff and change workflow files only after explicit confirmation.
Never edit automatically.
Options
| Option | Meaning |
|---|---|
--owner <owner> |
GitHub owner to scan (default: authenticated user) |
--repo <owner/repo> |
explicit repo (repeatable); disables owner scan |
--local <dir> |
scan local checkouts under <dir> instead of GitHub |
--tz <zone> |
local timezone for the Local column (default Europe/Zurich) |
--limit <n> |
max repos for the owner scan (default 200) |
--out <file> |
Markdown report (default target/cron-schedule-overview.md) |
--no-report |
console only, no report file |
--top <n> |
max redistribution suggestions (default 5) |
--queue-stats |
add observed queue time per UTC hour (via gh run list) |
--runs <n> |
max runs per repo for queue stats (default 50) |
Output
target/cron-schedule-overview.md:
- Scheduled workflows table: repo, workflow, cron (UTC), UTC time, local
time, frequency, intensity, evidence. - Runs per UTC hour histogram (bars).
- Runs per frequency / weekday.
- Collisions — UTC hours with two or more scheduled runs.
- Observed queue time per UTC hour (with
--queue-stats) — average and max
wait betweencreatedAtandstartedAtacross recent runs, as an empirical
contention signal. - Redistribution suggestions — high-intensity jobs in crowded hours moved
to empty ones; with--queue-statsthe empty hours with the lowest observed
queue are preferred.
Classification
Intensity is derived heuristically from the workflow content. See
references/classification-and-timezones.md for the exact patterns and the
UTC/timezone rules.
- high — Docker Compose, Testcontainers, service containers (
services:),
Kind/k3s cluster, matrix builds. - medium — Maven/Gradle full builds, Docker image builds.
- low — lint/format/validate and other meta workflows.
Examples
Example 1: Are my scheduled builds colliding?
User says: "Do my scheduled GitHub Actions builds collide?"
Actions:
- Run
bash scripts/cron-schedule-planner.sh --owner <owner> - Report the crowded UTC hours and the high-intensity jobs in them
- Point to
target/cron-schedule-overview.md
Result: Collision overview across all repos with concrete evidence.
Example 2: When should the Testcontainers build run?
User says: "When should I schedule the Testcontainers build?"
Actions:
- Run the planner and locate the crowded hours (collisions)
- Recommend one of the empty hours from the suggestions
- Offer the concrete cron diff, apply only after confirmation
Result: A low-contention slot and a ready-to-apply cron change.
Troubleshooting
gh: not found or authentication error
The script needs an authenticated gh. Run gh auth status; if that fails, the
user must authenticate. Alternatively use --local <dir> with local checkouts.
Empty report
No workflow in scope has an on.schedule entry. Check the repo scope
(--owner, --repo, --local) and that the workflows actually define
schedule: crons.
Times look off by one hour
Cron is always UTC. Local times are computed for the current date, so
daylight-saving transitions can shift them by one hour. Re-run after a DST
change or adjust manually.
--queue-stats shows little or no data
GitHub reports startedAt equal to createdAt for most runs (no recorded
queue), so the signal is sparse. Increase --runs to include older runs.
Treat large gaps with care: they can come from approval gates, concurrency
limits or manual runs, not only runner contention.





