# Mo Watchdog

> Observe one target or scan reachable Orca sessions for limits, overload, failures, questions, work, and completion. Use only when the user explicitly names Meta-O watchdog or mo-watchdog; nudge only an explicitly authorized exact target.

- Skill: `shkarupa-alex/mo-watchdog` (Agent Skill, multi-file: 4 files)
- Install (CLI): `npx skillmds@latest add shkarupa-alex/mo-watchdog`
- Raw SKILL.md: https://api.skillmd.com/api/skills/shkarupa-alex/mo-watchdog/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- License: MIT
- Author: shkarupa-alex (https://skillmd.com/u/shkarupa-alex)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/shkarupa-alex/mo-watchdog

---


# Watch backend sessions

Read [Watchdog behavior](references/watchdog.md) completely. Start only after the
user explicitly names Meta-O watchdog or `mo-watchdog`; a generic monitoring
question does not activate this skill.

Use `scripts/mo-watchdog.sh target --backend orca --session <id>` for one
session or `scripts/mo-watchdog.sh scan` for all reachable Orca sessions.
Observation is read-only. An explicit nudge additionally requires
`--nudge <message>` and exact target authorization; the script re-reads native
state and suppresses the nudge when that state changed. It reserves a bounded
private digest before delivery and suppresses the same message, an ambiguous
attempt, or a saturated unchanged state across later invocations. Nudges return
after native delivery; agent completion is observed separately.

Treat malformed or stale native state as a typed unsafe observation: report it,
fail closed, and do not nudge or guess the target state.

Do not inspect tracked project content or private provider state. Report the
native locator, classified state and action. Pattern misses are refined from
observed failures rather than hidden behind a support claim.