Poke#

When to Use#

Use poke for reminders, alarms, recurring nudges, deferred agent work, reply-driven escalation, and reminder side effects such as scripts that start before delivery or stop when the user replies.

Core Workflow#

1. Classify the poke style from the user's request.
2. Read the matching sub-skills/*.md file before building the command.
3. Check for an existing matching reminder before creating a duplicate.
4. Use --dry-run for unfamiliar flag combinations.
5. Forward plausible user replies with poke --reply; read sub-skills/replies.md first when handling replies.

Sub-skill Routing#

If making or handling...Read firstBasic one-shot, exact-time, recurring reminders, or deferred taskssub-skills/scheduling.md"Keep poking until I reply", escalation, rotating tones, heartbeat-style checks, unconfirmed followups, task intervalssub-skills/escalation.mdPre/post actions, side-effect scripts, wake lights, or stop-on-reply hookssub-skills/stages.mdUser replies such as done, cancel, snooze, later, or ambiguous inbound messagessub-skills/replies.mdQuiet hours, active hours, DND, urgent delivery, or "do not wake me" behaviorsub-skills/quiet-hours.mdMultiple channels, dependency chains, or per-channel visibilitysub-skills/multi-channel.mdListing, showing, cancelling, history, stats, presets, tones, path sets, or state layoutsub-skills/management.mdInferred promises like "I'll check on that tomorrow" without a precise fire timesub-skills/commitments.mdMigrating OpenClaw heartbeat polling into scheduled poke taskssub-skills/migration.md

Notes#

• --remind is for static reminder text.
• --task is for deferred agent work and needs --agent.
• --channel and --target identify where delivery and replies happen.
• Poke has no ears; inbound user messages must be forwarded explicitly.
• Command hooks ( --pre-cmd, --post-cmd, --on-pong-command, --if-unconfirmed-command) execute real local programs under your user account at scheduled time — only configure hooks with scripts you wrote or explicitly trust.