Agent Upgrade Kit
Drop-in upgrades for your OpenClaw agent. Makes it actually useful instead of just a chatbot.
Built by Dylan Essad. Battle-tested on a 24/7 Mac Mini running real business operations.
What This Is
Your OpenClaw agent out of the box is functional but basic. This kit adds:
- A personality framework that makes the agent direct, opinionated, and useful (not a sycophant)
- Persistent memory architecture that survives session restarts
- Skills that extend what the agent can do (quick notes, model switching, web scraping, research reports)
- Session startup rules so the agent reads its own context before responding
- A git-tracked workspace pattern so nothing gets lost
What This Is NOT
- Not a replacement for your existing OpenClaw setup
- Not business-specific (no client data, no credentials, no API keys)
- Not prescriptive about what you use it for
Install (Alongside Existing OpenClaw)
1. Clone this repo into your OpenClaw workspace
cd ~/.openclaw/workspace
git clone https://github.com/bigsouthballa/spencers-upgrade-kit.git _upgrade-kit
2. Copy the files you want
Don't blindly copy everything. Read each file, decide if it fits your setup.
Start with these (highest impact):
# The personality framework - makes the agent actually good
cp _upgrade-kit/SOUL.md ./SOUL.md
# Session startup rules - how the agent orients itself each session
cp _upgrade-kit/AGENTS.md ./AGENTS.md
# Identity template - fill in your agent's name and vibe
cp _upgrade-kit/IDENTITY.md ./IDENTITY.md
# Your profile - fill this in so the agent knows who you are
cp _upgrade-kit/USER.md ./USER.md
Then add skills:
# Copy all skills at once
cp -r _upgrade-kit/skills/* ./skills/
# Or pick individual ones:
cp -r _upgrade-kit/skills/model-switch ./skills/ # /opus /sonnet /haiku switching
cp -r _upgrade-kit/skills/btw ./skills/ # quick note capture
cp -r _upgrade-kit/skills/web-fetch ./skills/ # web scraping with Playwright
cp -r _upgrade-kit/skills/task-queue ./skills/ # async task acknowledgment pattern
Set up the brain and memory directories:
mkdir -p brain memory
cp _upgrade-kit/brain/IDEAS.md ./brain/IDEAS.md
cp _upgrade-kit/MEMORY.md ./MEMORY.md
cp _upgrade-kit/TOOLS.md ./TOOLS.md
cp _upgrade-kit/HEARTBEAT.md ./HEARTBEAT.md
3. Fill in YOUR details
Edit these files with your info:
IDENTITY.md- your agent's name, emoji, vibeUSER.md- your name, timezone, communication style, what you're working onTOOLS.md- your specific setup (SSH hosts, camera names, preferred voices, etc.)MEMORY.md- start empty, the agent will build this over time
4. Restart OpenClaw
openclaw gateway restart
Your agent will pick up the new files on the next session.
5. Clean up
rm -rf _upgrade-kit # remove the clone after copying what you need
What Each File Does
| File | Purpose |
|---|---|
SOUL.md | Personality, tone, communication rules. The agent reads this and becomes it. |
AGENTS.md | Session startup sequence: what to read, how to use memory, when to push to git. |
IDENTITY.md | Agent's name, emoji, avatar. Quick identity reference. |
USER.md | YOUR profile. Communication style, go/no-go signals, what motivates/frustrates you. |
MEMORY.md | Long-term memory. Agent writes here to remember things across sessions. |
TOOLS.md | Environment-specific notes (SSH hosts, API endpoints, device names). |
HEARTBEAT.md | Periodic check-in tasks. Leave empty if you don't want heartbeat actions. |
brain/IDEAS.md | Running idea log. The /btw skill appends here. |
Skills Included
| Skill | Command | What It Does |
|---|---|---|
| model-switch | /opus /sonnet /haiku | Switch between Claude models on the fly |
| btw | /btw [thought] | Capture a quick note without derailing the conversation |
| web-fetch | (agent uses automatically) | Scrape and extract content from web pages using Playwright |
| task-queue | (agent uses automatically) | Acknowledge long tasks immediately, notify when done |
Key Patterns Worth Understanding
Memory Architecture
MEMORY.md= long-term curated memory (agent reads on startup, writes when it learns something worth keeping)memory/YYYY-MM-DD.md= daily session logs (what happened today, decisions made, pending items)brain/= reference files (stack info, ideas, tools notes)
The agent wakes up fresh each session. These files ARE its memory. It reads them, updates them, and uses them to maintain continuity.
Research Reports
When you ask for a research report, the agent saves it as a markdown file in the workspace and pushes to git. Every report includes sources and citations. They accumulate and become a searchable knowledge base.
Git-Tracked Workspace
Everything the agent creates lives in the workspace and gets committed to git. Nothing is ephemeral. If the agent wrote it, you can find it in the repo history.
Tips
- Edit SOUL.md to match YOUR vibe. The default is direct and no-bullshit. If you want something different, change it.
- Fill in USER.md thoroughly. The more context the agent has about you, the less time you spend re-explaining things.
- Use /btw liberally. Ideas you capture in the moment compound over time.
- Check MEMORY.md occasionally. The agent curates its own memory but it's worth reviewing what it's keeping.
- The agent gets better the more you use it. Memory, skills, and context accumulate. Day 1 is the worst it'll ever be.
Credits
Built by Dylan Essad (@bigsouthballa). Framework developed through daily use running theFurnace.ai operations on OpenClaw.
No comments yet.