Local Cron Setup — Autonomous Runtime for school-core

For a fresh Hermes agent: clone this repo → run setup.sh → set env vars → optionally create the 2 local cron entries below → the framework runs autonomously.

What's in the repo vs. what's per-machine

Layer	Where	Who sets up	What it does
Cloud heartbeat	`school-loop.yml` (committed)	Clone + secrets	Runs the full pipeline every 5 min on the self-hosted Mac runner: fetch issues → classify → dispatch crews → CTO/COO review → PR → checkpoint state. No local cron needed for this.
Local digest	`scripts/school_daily_digest.py` (committed) + local crontab	User creates 1 crontab entry	Sends a daily Telegram digest of board state (NOW/NEXT/CUT lanes) at 9am weekdays.
Local health doc	`scripts/school_health_doc.py` (committed) + local crontab	User creates 1 crontab entry	Generates an interactive HTML health page at 8am weekdays with board state, failure classes, score trends, and clickable triggers.
Secrets + env	Local `.env` + credentials	User	OMNIROUTE_API_KEY, GH_TOKEN, AGENTMAIL_API_KEY, Telegram bot token.
Cron config	`~/.hermes/cron/jobs.json` (local-only, not in repo)	User	Hermes cron daemon config. The 2 entries below are examples.

Why only 2 local crons (not 5)

The 3 deleted entries (school-core (unified), agent-school-teacher-cto, agent-school-teacher-coo) were redundant with school-loop.yml:

school-loop.yml already runs issue_bridge --once every 5 min on the Mac self-hosted runner — the full pipeline including CTO/COO review.
Restoring school_cron.sh locally would double-execute the bridge every 5 min.
teacher_tick_cto.sh / teacher_tick_coo.sh ticked review separately, but the bridge handles CTO/COO review internally (reads cto_verdict/coo_verdict from task results).

Keep the 2 that do something the GitHub Action doesn't:

Daily digest — different channel (Telegram), different time (9am), different output (text summary vs HTML board).
Health doc — different output (interactive HTML with clickable triggers), different time (8am).

Setup steps

1. Clone + bootstrap

git clone https://github.com/branben/school-core.git
cd school-core
./setup.sh

2. Set env vars

# Required for the pipeline
export OMNIROUTE_API_KEY=<your-key>
export GH_TOKEN=<your-github-token>        # or set in .env
export AGENTMAIL_API_KEY=<your-key>        # optional, for alerts

# For the daily digest (Telegram)
export TELEGRAM_BOT_TOKEN=<your-bot-token>
export TELEGRAM_CHAT_ID=<your-chat-id>

See .env.example for the full contract.

3. Create the 2 local cron entries

The cron daemon is Hermes's built-in scheduler. Create entries via hermes cron:

# Daily digest — 9am weekdays, sends board summary to Telegram
hermes cron add \
  --name "school-daily-digest" \
  --schedule "0 9 * * 1-5" \
  --script "python3 /path/to/school-core/scripts/school_daily_digest.py" \
  --workdir "/path/to/school-core" \
  --no-agent

# Health doc — 8am weekdays, generates interactive HTML
hermes cron add \
  --name "school-health-doc" \
  --schedule "0 8 * * 1-5" \
  --script "python3 /path/to/school-core/scripts/school_health_doc.py" \
  --workdir "/path/to/school-core" \
  --no-agent

Or edit ~/.hermes/cron/jobs.json directly — the schema is visible in the existing entries.

4. Verify the pipeline works

cd school-core
nix develop .#verifyShell   # Tier A required
python issue_bridge.py --repo branben/sound-royale-ny --once

5. (Optional) Point at a different repo

The SCHOOL_REPO env var in school-loop.yml controls the target. Change it to point the framework at any repo:

env:
  SCHOOL_REPO: "owner/your-repo"

The local digest and health doc scripts read data/board.json — they're repo-agnostic as long as the board is populated.

What a fresh agent user gets from the repo

school-loop.yml — the 5-min heartbeat (no setup beyond secrets)
scripts/school_daily_digest.py — the digest script (crontab entry is per-machine)
scripts/school_health_doc.py — the health doc script (crontab entry is per-machine)
docs/setup.md — Tier A/B setup (Nix, verify shell, pipeline run)
docs/sdlc/ — the full SDLC framework (loops, phase-map, context-skills)

What a fresh agent user must set up per-machine:

Secrets (.env + credentials)
Local cron entries for digest + health doc (if they want them)
TELEGRAM_BOT_TOKEN / TELEGRAM_CHAT_ID for the digest

Deleting the redundant crons

If you have the 3 old entries still in ~/.hermes/cron/jobs.json, remove them:

hermes cron remove --id <school-core-unified-id>
hermes cron remove --id <teacher-cto-id>
hermes cron remove --id <teacher-coo-id>

Or edit jobs.json directly and remove the entries with those IDs.

Design rationale

The framework is repo-agnostic: the pipeline runs against whatever SCHOOL_REPO points at. The local crons are delivery-specific: they send digests and generate health docs for the operator's machine. Separating them means:

A new user clones the repo and gets the pipeline + scripts immediately.
They opt into the local delivery layer (digest + health doc) with 2 crontab entries.
The cloud heartbeat (school-loop.yml) works without any local cron.

This is the split recommended by omh-automation-blueprint: host automation (crontabs) stays on the operator's machine; the framework ships the scripts + a setup note.