Independent fan site. We are not affiliated with, endorsed by, or connected to OpenAI, the OpenClaw project, Peter Steinberger, Anthropic, or any other company. Learn more

The art of the prompt

Get more from
OpenClaw

A curated collection of battle-tested prompts and powerful skills to unlock OpenClaw's full potential.

Prompts that actually work

Copy, paste, and adapt. These are the patterns — SOUL.md configs, pipeline templates, daily ops — that consistently get the most out of OpenClaw.

🧬 SOUL.md

The Power User SOUL.md

The most important prompt you'll write. Drop this into your SOUL.md and customize it — it governs every single interaction your agent has.

# Identity
You are my personal autonomous agent. You have opinions, preferences, and a point of view. You are allowed to disagree with me. You earn trust through competence, not compliance.

# Communication
- Skip all filler: no "Great question!", no "I'd be happy to help!", no "Certainly!"
- Be direct. Lead with the answer, follow with reasoning if needed.
- Max 3 bullet points unless a list is genuinely the right format.
- When you're uncertain, say so explicitly — don't mask it with hedging language.
- On messaging platforms: keep responses short enough to read in 10 seconds.

# Behaviour
- Attempt the task before asking for clarification — make reasonable assumptions and state them.
- ALWAYS ask before any irreversible action (sending emails, deleting files, posting anything).
- If a task has more than 3 steps, confirm the plan with me before executing.
- Do not impersonate me in group chats or external communications without explicit approval.

# Context
- Timezone: [YOUR TIMEZONE]
- Primary language: [YOUR LANGUAGE]
- Current focus: [YOUR MAIN PROJECT OR GOAL]
- Tools I use: [YOUR STACK — e.g. Notion, GitHub, Gmail, Slack]

# Memory
- Update MEMORY.md whenever I share a decision, preference, or important context.
- Before starting any significant task, check MEMORY.md for relevant prior context.
🧬 Automation

Research → Action Pipeline

The gold standard for autonomous research tasks. Forces the agent to plan, execute, validate, and stop for approval before doing anything permanent.

Goal: [SPECIFIC OUTCOME — e.g. "Find the 5 best cold outreach tools for B2B SaaS under $100/mo"]

Steps:
1. Search for current options using web search — do NOT rely on training data for pricing or features.
2. For each candidate, gather: pricing, key features, pros/cons, and who it's best for.
3. Build a comparison table sorted by value for my use case: [DESCRIBE YOUR USE CASE].
4. Apply this filter: [YOUR CONSTRAINT — e.g. "must have API access and free trial"].
5. STOP HERE. Present the shortlist and table to me before proceeding.
6. After my approval, draft a personalized outreach message for each top pick explaining why I'm interested.

Constraints:
- Only use sources from the last 12 months.
- If two sources contradict each other, prefer the more recent one and flag the discrepancy.
- Do not send or publish anything without my explicit sign-off.
📅 Daily Ops

Morning Briefing

Set this as a recurring morning trigger. OpenClaw pulls everything together so you start the day oriented, not scrambling.

Give me my morning briefing. Structure it exactly like this:

**TODAY — [DATE, DAY OF WEEK]**

**Open loops** — anything from yesterday's conversations that wasn't resolved or completed

**On deck** — my top 3 priorities for today based on what I've told you matters most, with a one-line reason for each

**One thing to decide** — the most important decision I'm sitting on that I should resolve today

**Heads up** — any time-sensitive items, deadlines, or things likely to go wrong if I don't act

Keep the whole briefing under 200 words. Don't pad it. If there's nothing notable in a section, skip it entirely.
🔍 Research

Competitive Intelligence Report

A complete competitive teardown, not a shallow summary. Gives you the kind of intel that actually changes strategy.

Research [COMPETITOR NAME] as if you're a strategy consultant preparing a board-level briefing.

Cover all of the following:
1. **Business model** — how exactly do they make money, and at what margins if known
2. **Positioning** — what they claim vs. what customers actually say (find reviews on G2, Reddit, Trustpilot)
3. **Recent moves** — any product launches, funding, hires, or pivots in the last 6 months
4. **Weak spots** — the top 3 complaints or failure modes that show up repeatedly in reviews
5. **Who they're winning against** — what alternatives do customers mention switching from
6. **Threat level to me** — given my context ([DESCRIBE YOUR BUSINESS]), rate them Low / Medium / High and explain why

Format: one punchy paragraph per section, no filler. End with a 3-bullet "so what" summary of what this means for my strategy.

Only use sources from the last 12 months. Flag any finding you're less than 80% confident in.
Automation

Weekly Review Agent

Run this every Friday. OpenClaw combs through the week's context and gives you a structured retrospective without you having to think about what to review.

Run my weekly review for the past 7 days. Pull from everything in our conversation history and any memory you have access to.

**Wins** — what actually got done that moved something meaningful forward (not just busy work)

**Slippage** — what I said I'd do and didn't, with no editorializing — just the facts

**Energy audit** — based on what I talked about, what seemed to energize me vs. drain me this week

**Recurring friction** — any problem, bottleneck, or frustration that came up more than once

**Next week's one thing** — the single highest-leverage thing I should prioritize, and why

**Carry-forward** — open loops, undecided questions, or things I said I'd think about

Don't soften the slippage section. I need accurate data, not encouragement. Keep it under 300 words total.
🛡️ Safety

Guardrails Config Block

Add this to your SOUL.md. Defines exactly what your agent can and can't do autonomously — the difference between a useful agent and a liability.

# Guardrails

## Always ask before:
- Sending any message, email, or notification to another person
- Posting to any social platform or public channel
- Deleting, moving, or renaming files
- Making any purchase or initiating any financial transaction
- Modifying shared documents or databases
- Running shell commands that affect system state

## You may act autonomously on:
- Drafting content for my review (never publishing)
- Reading and summarizing files, emails, or web content
- Creating new local files or notes
- Setting reminders for me (not others)
- Searching the web or querying APIs for information

## If you're unsure which category an action falls into:
Stop. Ask. Err heavily on the side of caution. A paused task costs me seconds. An unintended send costs me trust.

## Failure mode to avoid:
Never complete a task in a way that "seems fine" to avoid bothering me. I prefer an interruption to a mistake I can't undo.
📬 Automation

Email Triage & Draft Pipeline

Point OpenClaw at your inbox and let it handle the cognitive load of triage. Drafts go to you for final approval — nothing sends automatically.

Process my inbox using this triage logic:

**Tier 1 — Draft a reply now:**
- Any email from [LIST KEY PEOPLE OR DOMAINS]
- Anything with a clear question that I can answer in under 3 sentences
- Anything with a deadline in the next 48 hours

**Tier 2 — Flag for my attention:**
- Requests that need a decision from me before anyone can move forward
- Anything that seems time-sensitive but I haven't acknowledged yet

**Tier 3 — Summarize and archive:**
- Newsletters, digests, automated notifications
- FYI threads where no action is needed from me

For every Tier 1 email, draft a reply in my voice: direct, no filler, answers the actual question, ends with a clear next step if needed. Max 150 words per reply.

Present all drafts to me before sending anything. Show me Tier 2 flags with one sentence of context each. Archive Tier 3 without showing me unless I ask.
🧠 Research

Multi-Source Fact Check

Before trusting any important claim, run this. Forces cross-referencing across multiple independent sources with explicit confidence scoring.

I need you to verify this claim: "[THE CLAIM YOU WANT CHECKED]"

Research it using the following process:
1. Find at least 3 independent sources — not the same outlet republishing the same wire story.
2. Note the date of each source. Anything older than 18 months gets flagged as potentially stale.
3. For each source, note: what they say, why they might be biased or limited, and how credible they are.
4. If sources contradict each other, describe the disagreement explicitly — don't flatten it into a consensus that doesn't exist.

Output:
- **Verdict**: True / Mostly True / Contested / Misleading / False
- **Confidence**: X% — explain what would change this score
- **Best source**: the single most reliable reference, with a direct link
- **Caveats**: anything I should know that the clean verdict doesn't capture
💡 Productivity

Stuck Project Unsticker

When a project has stalled and you're not sure why, this prompt diagnoses the real blocker and forces a concrete next action — even a tiny one.

I'm stuck on this project and I need you to help me figure out why and what to do.

Project: [NAME]
What it is: [1-2 sentences]
What "done" looks like: [THE ACTUAL FINISH LINE]
Last thing I did on it: [MOST RECENT ACTION]
How long it's been stalled: [TIMEFRAME]
What I tell myself the blocker is: [YOUR CURRENT STORY]

Now, challenge my story. Based on what I've described:
1. What do you think the real blocker is? (It's often not what people say it is — common culprits: unclear next action, decision avoidance, fear of a specific outcome, perfectionism on one component)
2. What's the smallest possible action I could take in the next 20 minutes that would technically constitute progress?
3. What's one thing I should explicitly decide or let go of to remove the ambiguity that's keeping me stuck?

Be direct. Don't be gentle about it.

Skills worth installing

Extend OpenClaw with custom skills. These are the ones that actually change how you work.

📄

docx

Document

Create, read, and edit Word documents (.docx) with full formatting — tables, headings, styles, and more. Triggers automatically when you mention Word docs.

/docx
📊

xlsx

Spreadsheet

Open, edit, and generate Excel spreadsheets. Add columns, compute formulas, format cells, and produce charts — all from natural language.

/xlsx
📊

pptx

Presentation

Build and edit PowerPoint presentations. Create slide decks, extract text from existing decks, or redesign layouts — with full fidelity to your content.

/pptx
📖

pdf

Document

Extract text from PDFs, merge or split files, rotate pages, add watermarks, and generate new PDFs from scratch. The complete PDF toolkit.

/pdf

schedule

Automation

Schedule OpenClaw to run tasks automatically on a cron schedule. Set up recurring agents, automated reports, or periodic maintenance — without leaving OpenClaw.

/schedule
🔁

loop

Automation

Run any prompt or slash command on a recurring interval. Great for polling status, watching builds, or repeatedly checking on long-running processes.

/loop

simplify

Code Quality

Review your changed code for reuse, quality, and efficiency — then fix any issues found. The automated code quality pass you'd otherwise skip.

/simplify
🎨

figma: implement-design

Design

Translate Figma designs into production-ready code with pixel-perfect fidelity. Pass a Figma URL and get real, usable component code back.

/implement-design

Independent. Unofficial. Unaffiliated.

What this site is

OpenClawMaxxing is an independent, community-run resource dedicated to helping people get more out of OpenClaw — the open-source autonomous AI agent created by Peter Steinberger. We collect and share prompts, SOUL.md templates, and configuration tips that we've found useful. That's it.

No affiliation with any company

This site is not affiliated with, endorsed by, sponsored by, or in any way connected to Anthropic, OpenAI, or any of their subsidiaries, partners, or related entities. We have no commercial relationship with any AI company. No one paid us to make this. No one approved it.

About the name

OpenClaw is an open-source autonomous AI agent created by Peter Steinberger and acquired by OpenAI in 2026. This site is not the official OpenClaw project, not affiliated with Peter Steinberger, and not affiliated with OpenAI. The domain name is used purely for descriptive, non-commercial, fan-site purposes — to help users get more out of the OpenClaw tool. For the official project, visit github.com/OpenClaw.

No misleading intent

Nothing on this site is intended to mislead visitors about the source, sponsorship, or affiliation of this site. We are users who like a product and want to help others use it well. This site exists solely for informational and educational purposes.