Cron Fleet — Multi-Profile Scheduled Jobs

Drive hermes cron across all profiles in a Hermes installation. This skill
covers fleet-wide audits, the self-heal pattern, and profile-scoped job management.

Base CLI lives in the bundled hermes-agent skill (CLI Reference → Cron Jobs).
This skill adds the multi-profile fleet patterns and watchdog workflow.

Quick Ref

hermes cron list            # active jobs only
hermes cron list --all      # includes disabled / completed jobs
hermes cron resume <id>     # re-enable a disabled job
hermes cron pause <id>      # disable a job
hermes cron run <id>        # trigger next tick immediately
hermes cron remove <id>     # delete permanently

Invocation note. There is no standalone cronjob binary. All cron
operations go through the hermes cron CLI (from the terminal tool,
or via the cronjob toolset in live sessions that expose it). Set
HERMES_HOME to target a specific profile.

Profiles & Fleet Context

Each Hermes profile has its own $HERMES_HOME and therefore its own cron
fleet. To manage a profile from the outside:

export HERMES_HOME=/home/sakthai/.hermes/profiles/<name>
hermes cron list --all

Common profiles on this machine:

• default (~/.hermes) — Hermes bot
• sakthai (~/.hermes/profiles/sakthai) — Saksee bot
• hermesagent (~/.hermes/profiles/hermesagent) — SakThai bot
• saksit (~/.hermes/profiles/saksit) — SakSit bot

Fleet Watchdog Pattern

When asked to audit the entire fleet:

1. Enumerate every profile. Check each known $HERMES_HOME.
2. List all jobs with hermes cron list --all per profile.
3. Self-heal disabled-ok jobs. In JSON, enabled: false +
   last_status: "ok" means the job was healthy before being paused.
   In CLI, these may appear as [completed]. Re-enable with:

   hermes cron resume <job_id>

4. Report chronic errors. Jobs with last_status: "error" or a
   non-null last_delivery_error need human review. Do NOT auto-re-enable
   these — output the job ID, name, and the error signal.
5. Silent when healthy. If every profile is green and no actions were
   taken, emit exactly [SILENT] to suppress delivery.

State Quirks

JSON field	CLI display	Notes
enabled: false	[completed]	Completed one-shots or paused jobs
enabled: true	[active]	Running normally
last_status: "ok"	ok	Last run succeeded
last_status: "error"	error	Last run failed
non-null last_delivery_error	(visible in JSON)	Delivery failure — separate from run status
last_status: null	(not shown)	Job never ran (newly created) — not an error

hermes cron list (no --all) filters to [active] jobs only. Use --all
to include disabled and completed jobs.

Self-Heal Watchdog Template

When asked to create a self-healing watchdog, use this prompt for a cron
job that runs every 5m:

You are a cron fleet watchdog. Every tick:

1. Run `hermes cron list --all` across every known profile and inspect all jobs.
   - If you need richer metadata (last_delivery_error, exact timestamps), fall
     back to the Python API: `python3 -c "from cron.jobs import list_jobs; print(list_jobs(include_disabled=True))"`.
2. For any job that is `enabled: false` AND has `last_status: 'ok'`
   (i.e., it was working before being disabled), re-enable it:
   `hermes cron resume <job_id>`.
3. For jobs with `last_status: 'error'` or a non-null delivery error,
   **do NOT** auto-re-enable. Instead, emit a concise report listing
   the job ID, name, and the error signal.
4. Output `[SILENT]` when no actions were taken and the fleet is healthy.

Key constraints:

• Schedule: every 5m, repeat: forever.
• The watchdog must stay quiet when there is nothing to report.
• Do not recursively schedule more cron jobs from inside the watchdog.
• There is no standalone cronjob binary; all cron operations go through the
  hermes cron CLI.
• hermes cron list (no --all) filters to [active] jobs only. Use --all
  to include disabled and completed jobs.
• The CLI display omits last_delivery_error; to inspect it, read
  $HERMES_HOME/cron/jobs.json directly or use the Python API above.

Repo / Source

Source lives at ~/.hermes/hermes-agent/cron/. The JSON store is at
$HERMES_HOME/cron/jobs.json.

References

• Fleet watchdog pattern & cron cross-profile notes: references/fleet-watchdog.md
• Source code layout: ~/.hermes/hermes-agent/cron/