SKILL: smart-wake

Objective

Standard mechanism:

1. 1. Auto-save state before timeout / reaching timeout threshold.
2. Cron wake to re-invoke agent at scheduled time.
3. Resume task from latest checkpoint using sessions_spawn (or session_spawn depending on runtime) with appropriate wakeMode.

Activation Triggers

• - Task duration exceeds default subagent timeout.
• Requires periodic polling (by minute/hour) to continue processing.
• Multi-step workflow that may pause mid-execution but must resume accurately.
• Running overnight / off-hours but still needs to auto-continue.

Available Tools (Built-in)

• - sessions_spawn / session_spawn (spawn new work session)
• Gateway cron jobs (schedule wake events)
• Memory/slot for state persistence (state checkpoint)

Do not create new tools. Only use existing mechanisms.

State Contract (MANDATORY)

CODEBLOCK0

Requirements:

• - Update checkpoint after each completed step.
• Final checkpoint before timeout.
• Resume always reads latest checkpoint; never re-run blindly.

Standard Workflow

Step 1 — Preflight (Estimate Timeout)

• - Determine subagent timeout and total task duration.
• If estimated_duration > 70% timeout: enable smart-wake immediately.
• Create task_id and initial checkpoint.

Step 2 — Execute by Small Chunks

• - Divide task into 5–15 minute chunks.
• After each chunk:

Step 3 — Pre-timeout Auto-save

• - When ~10–20% timeout remains:

Step 4 — Schedule Cron Wake

• - Register Gateway cron job to wake after appropriate interval (e.g., 2–10 minutes depending on load).
• Cron payload MUST contain:

Step 5 — Spawn Resume Session

• - At wake time, call sessions_spawn (or session_spawn) with wakeMode enabling auto-resume via cron.
• Pass concise context:

Step 6 — Resume + Idempotency Guard

• - New session must:

• - When entire task completes: set status = done, cancel remaining cron wakes.

Sample Resume Packet (Reference)

CODEBLOCK1

Note: Field names may vary by Gateway/OpenClaw version, but checkpoint + wakeMode + resume context principles are mandatory.

Operational Rules

1. 1. Never resume without valid checkpoint.
2. Never overwrite artifacts without checksum/timestamp verification.
3. No infinite wake loops: limit retry_count (e.g., max 5).
4. Every wake must have clear reason (timeout_recovery, dependency_ready, scheduled_progress).
5. Clean up on done: completed task must clear related cron jobs.

Output Format After Each Wake Cycle

CODEBLOCK2

Anti-patterns (FORBIDDEN)

• - Running long tasks without periodic checkpointing.
• Waking without passing task_id/resume context.
• Re-running entire pipeline from start when only 1 step is missing.
• Not canceling cron after done.

Expected Outcomes

• - No state loss on subagent timeout.
• Long tasks progress steadily via cron wake.
• Full system auditability: checkpoint → wake → resume → done.