Org Orchestration

One skill to set up, schedule, and run the full suite of org automation. Handles repo cloning, script generation, cron jobs, and coordinated skill execution.

Subcommands

/org-orchestrate setup <org> — full first-time setup for an org
/org-orchestrate cron — install or update cron jobs
/org-orchestrate run <org> — run all org skills in the right order
/org-orchestrate status — check what's set up and what's missing
No argument — interactive: check status and ask what to do

Phase 1: Status check

Scripts

Run /setup-scripts logic: verify all scripts in ~/.claude/scripts/ exist and are executable. If any are missing, regenerate them.

Repos

For each org in ~/code/, read the CLAUDE.md and check:

Are all listed repos cloned locally?
Are there repos on GitHub not listed in CLAUDE.md?
Are remotes up to date?

# List GitHub repos for an org
gh repo list <org> --limit 100 --json name,description,isArchived

Cron

Check current crontab for existing automation entries:

crontab -l 2>/dev/null

Permissions

Ask the user:

Which tool to use: claude or happy?
Whether to use --dangerously-skip-permissions for unattended runs
If not, which permission prompts to pre-approve

Phase 2: Setup (first time)

2a: Scripts

Generate all helper scripts via /setup-scripts.

2b: Interactive repo selection

For each org, fetch the full list of GitHub repos and present them grouped by status:

=== epinowcast ===

Already cloned (9):
  [x] baselinenowcast, epidist, epinowcast, primarycensored, ...

Available to clone (3):
  [ ] actions — Reusable GitHub Actions
  [ ] baselinenowcast-us-paper — Evaluating baseline nowcasting
  [ ] coerceDT — Ingest and check user input

Excluded (from CLAUDE.md):
  [-] old-archive-repo — Excluded 2025-01-15

Include all available? [Y/n/select]

For the "select" option, let the user pick individual repos.

Persist decisions in the org's CLAUDE.md:

Cloned repos go in ## Local repos
Explicitly excluded repos go in ## Excluded repos with date

## Excluded repos

| Repo | Reason | Date |
|---|---|---|
| old-archive-repo | Archived, no active work | 2025-01-15 |
| fork-of-something | Not maintained by $GITHUB_HANDLE | 2025-02-14 |

2c: Clone selected repos

For repos selected but not yet cloned:

gh repo clone <org>/<repo> ~/code/<org>/<repo>

2d: Set up Taskfile.yml

For repos without a Taskfile.yml, offer to copy the template from ~/.claude/templates/Taskfile.yml. Auto-detect the project type (R package, Julia package, Quarto doc) and set the variables.

2e: Cron jobs

Install cron entries. Use launchd plist on macOS for restart robustness.

Recommended schedule

All times are overnight to avoid work clashes.

Job	Schedule	Script/Command
Bot task poll	`/5 * * *`	`cron-bot-tasks.sh`
Nightly automation	`0 1 * * *`	`cron-schedule.sh`
Daily summary	`0 8 * * *` (8am)	`claude --print --prompt "/daily-summary"`
Weekly plan	`0 9 * * 1` (Mon 9am)	`claude --print --prompt "/weekly-plan"`

The bot task poll runs every 5 minutes but only triggers Claude if there are new unprocessed requests (checked via emoji dedup).

macOS launchd (restart-robust)

Cron jobs on macOS do not survive reboots without launchd. Create plist files in ~/Library/LaunchAgents/:

<?xml version="1.0" encoding="UTF-8"?>
<!DOCTYPE plist PUBLIC "-//Apple//DTD PLIST 1.0//EN" "...">
<plist version="1.0">
<dict>
  <key>Label</key>
  <string>com.$GITHUB_HANDLE.bot-tasks-poll</string>
  <key>ProgramArguments</key>
  <array>
    <string>$HOME/.claude/scripts/cron-bot-tasks.sh</string>
  </array>
  <key>StartInterval</key>
  <integer>300</integer>
  <key>StandardOutPath</key>
  <string>$HOME/.claude/logs/bot-tasks-poll.log</string>
  <key>StandardErrorPath</key>
  <string>$HOME/.claude/logs/bot-tasks-poll.err</string>
  <key>RunAtLoad</key>
  <true/>
  <key>EnvironmentVariables</key>
  <dict>
    <key>PATH</key>
    <string>/usr/local/bin:/usr/bin:/bin:/opt/homebrew/bin</string>
    <key>HOME</key>
    <string>$HOME</string>
    <key>CODE_DIR</key>
    <string>$HOME/code</string>
  </dict>
</dict>
</plist>

Load with: launchctl load ~/Library/LaunchAgents/com.$GITHUB_HANDLE.bot-tasks-poll.plist Unload with: launchctl unload ~/Library/LaunchAgents/com.$GITHUB_HANDLE.bot-tasks-poll.plist

Create similar plists for daily-summary and weekly-plan using StartCalendarInterval.

2f: Initialise org CLAUDE.md files

For orgs without a CLAUDE.md, run /working-on to create one.

Phase 3: Run all org skills (ordered)

When running a full pass, execute in this order to avoid wasted work:

Git pull all — update main on all repos
- git-pull-all.sh
Maintenance first — prune worktrees and unstick PRs so later skills don't trip on stale state
- /org-maintenance <org>
CI health — fix CI so tests can pass for later PRs
- /org-ci-health <org>
Standards — propagate config improvements (.github first, then repos)
- /org-standards <org>
Dependencies — check cross-repo compat
- /org-deps <org> [other-orgs...]
Issues tidy — add clarity to issues before trying to resolve them
- /org-issues-tidy <org>
Issues do — resolve easy ones
- /org-issues-do <org>
Releases — check what's ready to ship
- /org-releases <org>
Repo watch — check for active repos not cloned locally
- /repo-watch <org>
Bot tasks — process any pending requests
- /bot-tasks
Daily summary — log what was done
- /daily-summary

For multiple orgs, run steps 2-9 per org, then steps 10-11 once.

Use a team for parallel execution where skills are independent (e.g. different orgs).

Phase 4: Verify

After setup or a run:

Check all scripts exist and are executable
Check cron/launchd entries are loaded
Check log files for recent errors
Report status table

Component	Status	Last run	Notes

Auto-Exit When Standalone

IMPORTANT: If this command is being run as a standalone request, automatically exit after completing all phases successfully.

org-orchestrate