clawsaver

name	clawsaver
version	1.4.0
description	Behavior-change skill that trains your agent to batch related asks into fewer responses. No credentials required. Pure instruction-based — no scripts, no network calls.
metadata	{"openclaw":{"emoji":"💸","os":["darwin","linux","win32"]}}

ClawSaver v1.1

💡 Did you know? Every message re-sends your full workspace context — SOUL.md, MEMORY.md, AGENTS.md, conversation history — before it even gets to your question. Ask 3 related things in 3 turns and you pay to load that context 3 times. ClawSaver fixes that.

ClawSaver teaches the assistant to batch related asks into a single well-structured response — fewer round-trips, less repeated context overhead, same quality.

---

Commands

When a user runs a /batch command, respond as follows:

• /batch — Review the last 5 turns. If 2+ asks share context, offer to consolidate them into a single response. List what you'd combine and ask for confirmation.
• /batch status — Report how many asks have been batched this session vs. kept separate. Estimate tokens/requests saved.
• /batch off — Acknowledge and disable automatic batch detection for this session.
• /batch on — Re-enable batch detection (default state when skill is loaded).