feat(wacli): add stateful monitor check helper

[artemgetmann]

Review Fast Path

• User path fixed: agent asks to monitor a WhatsApp contact and now has one stable helper command for stateful polling instead of hand-writing wacli-recent-reply + ad hoc state-file choreography.
• Proof: pnpm vitest run skills/wacli/scripts/wacli-monitor-check.test.ts src/whatsapp/wacli-monitor-check.test.ts src/agents/system-prompt.test.ts -> 2 files, 48 tests passed.
• Shared-state footgun removed: monitor jobs no longer need the model to invent a durable state-file path on every run; the helper derives a stable per-target state file under ~/.openclaw/wacli-monitor-state.
• Still hurts: this is stacked on top of fix(wacli): persist monitor baseline state #411 and does not itself prove full autonomous WhatsApp auto-send.

Why This Matters

• The monitor bug fix in fix(wacli): persist monitor baseline state #411 works, but the generic agent path is still prompt-fragile if the model has to reconstruct the shell recipe.
• This patch gives the agent one first-class helper for the monitoring use case and updates the guidance to point at it.

Scope Boundary

• Added skills/wacli/scripts/wacli-monitor-check.{ts,sh}.
• Added focused tests for derived state-file behavior and persisted first-run baseline.
• Updated skills/wacli/SKILL.md and src/agents/system-prompt.ts guidance.
• Did not change reply policy, OAuth/profile healing, or the monitor bootstrap logic already fixed in fix(wacli): persist monitor baseline state #411.

Verification

• pnpm vitest run skills/wacli/scripts/wacli-monitor-check.test.ts src/whatsapp/wacli-monitor-check.test.ts src/agents/system-prompt.test.ts
• Result: 2 files, 48 tests passed

AI Assistance

• AI-assisted
• Testing degree: targeted