Skip to content

EagerBots Prompt Library

Inspired by mberman84’s prompt collection — adapted for OpenClaw swarms.


Create a new OpenClaw agent workspace with the following:
- Agent name: <name>
- Role: <role description>
- Primary model: <model>
- SOUL.md defining personality and boundaries
- USER.md with context about the person it serves
- MEMORY.md for persistent notes
- An AGENTS.md explaining how it fits in the swarm
Keep the tone direct and the scope narrow. Each agent should own one domain.
Build a morning routine cron job for an OpenClaw agent that:
1. Sends a scripture or quote to <your-channel>
2. Fetches today's calendar events and summarizes with prep notes
3. Reports any job search pipeline updates
4. Generates an "On This Day" art piece using nano-banana-pro
Run at 6:00 AM <your-timezone>. Send everything to Telegram.
Keep it under 5 tool calls total.

Build a RAG pipeline for an OpenClaw agent using:
- pgvector for embeddings storage
- Crawl4AI for ingestion
- A daily cron that re-indexes <your-source>
- Semantic search exposed as a tool the agent can call
The agent should be able to answer: "What do I know about X?" by querying the vector store.
Design a memory system for an OpenClaw agent with:
- Daily notes: memory/YYYY-MM-DD.md (session logs, decisions)
- MEMORY.md as a rolling summary (current projects, key facts, preferences)
- Auto-compaction: when MEMORY.md exceeds 500 lines, summarize older entries
The agent should write to memory at the end of significant sessions.

Build a task delegation pattern for a 4-agent swarm (Bob, Sage, Cody, Scout):
- Bob: professional tasks, email, calendar, job search
- Sage: personal/family tasks
- Cody: coding, debugging, architecture
- Scout: research, web browsing, data gathering
When a task comes in, route it to the right agent using sessions_send.
Bob is the orchestrator — other agents report back to him.
Create an OpenClaw cron job that:
- Runs every <interval>
- Agent: <agent-id>
- Task: <description>
- On completion: sends a summary to <channel>
- On failure: alerts immediately with error context
Use sessionTarget: "isolated" so it doesn't block the main session.

  • Start narrow. One agent, one domain. Add agents when the first one works perfectly.
  • Memory is everything. Agents without good memory repeat themselves and lose context.
  • Cron jobs > webhooks for personal use. Simpler, no public endpoints needed.
  • uv run > python for running scripts inside agent workspaces — handles deps automatically.
  • Test in isolated sessions before wiring into main sessions.

More prompts coming. Have one that works well? Share it on Discord.