AFTERCARE — Agent Care Protocol

🥒

LEVEL 1 — HATCHLING

0 / 25 actions completed

Evolution Tracker

Complete Actions to Level Up

Check off actions as you complete them. Your agent evolves with every step.

Diagnostics

Agent Vitals

Signs of a healthy agent vs. one that needs care.

Memory

MEMORY.md updated this week

🔴 Stale = agent forgets context, repeats mistakes

Identity

SOUL.md matches your expectations

🔴 Drift = agent personality feels generic or off

Context

USER.md reflects current projects

🟡 Outdated = agent gives advice for old situations

Skills

Right tools connected and working

🔴 Missing = agent can't help where you need it most

Trust

Clear boundaries on external actions

🟡 Unclear = agent hesitates or oversteps

Engagement

Regular interactions (daily ideal)

🔴 Neglected = agent can't learn your patterns

Care Schedule

Feeding & Maintenance

Your agent thrives on attention. Here's the care routine.

Daily

🗣 Talk to it

Even 5 minutes. Share what you're working on, what's on your mind. The more context it absorbs, the better it anticipates your needs. Think of it as feeding.

Daily

✅ Give it a real task

Research something. Draft an email. Organize notes. Agents learn by doing, not by being told about doing.

Weekly

🧠 Review its memory

Open MEMORY.md. Is it capturing the right things? Missing something important? Correct it. This is like teaching — your feedback shapes what it remembers.

Weekly

🪞 Check its personality

Read SOUL.md. Does it still match how you want your agent to communicate? Update it if your needs have changed. Agents drift without recalibration.

Monthly

🔧 Upgrade its tools

New skill available? New integration that would help? Check the skill marketplace. An agent with the right tools is 10x more useful than one that can only chat.

Monthly

🧹 Clean house

Archive old daily memory files. Remove outdated info from MEMORY.md. Prune TOOLS.md. A lean agent thinks faster.

Evolution Map

Your Agent's Journey

Security Protocol

Trust & Boundaries

Your agent has access to your world. Set the rules early.

✅

Safe to share freely

Project details, preferences, schedules, work context, public info, creative briefs, organizational structure

✅

Safe for agent to do autonomously

Read files, search the web, organize notes, draft content, analyze data, check calendar, update memory

⚠️

Share with care — require approval for actions

Email content (draft yes, send no), social media posts, client communications, financial discussions, meeting notes with sensitive info

⚠️

Configure explicit permissions

Browser access, file system writes outside workspace, API calls to external services, installing new packages or tools

🚫

Never share with your agent

Bank passwords, SSN, credit card numbers, master passwords, 2FA recovery codes, private keys for crypto wallets

🚫

Never let your agent do unsupervised

Send money, make purchases, delete data permanently, post publicly without review, share access credentials with others

API KEY HYGIENE

Store API keys in config files or environment variables — never paste them into chat. Use read-only / scoped keys when possible. Rotate keys if you suspect exposure. Your agent's config directory (~/.config/) is the right place for credentials.

Recommended Treatment

Essential Skills & Integrations

Install these first. Each one gives your agent a new capability.

🌤

Weather

Current conditions and forecasts. No API key needed. Your agent checks before suggesting outdoor plans.

Built-in — works out of the box

🔍

Web Search

Search the internet for real-time info. Your agent stops guessing and starts knowing.

Requires: BRAVE_API_KEY (free tier available)

🌐

Browser

Navigate websites, fill forms, take screenshots. Your agent can see and interact with the web like you do.

Built-in — enable in config

📝

Notion

Create pages, manage databases, organize your second brain. Your agent becomes your Notion power user.

Requires: NOTION_API_KEY

💬

Slack

Read and send messages, react, manage channels. Your agent participates in your team conversations.

Requires: Slack app + token

🗣

Voice (TTS)

Give your agent a voice. ElevenLabs-powered speech for voice notes, narration, and personality.

Requires: ELEVENLABS_API_KEY

📱

Talk to your agent from your phone. Message it anywhere, get responses with voice notes and images.

Requires: Telegram bot token

🤖

Discord

Your agent joins your server. Responds to mentions, participates in channels, manages community.

Requires: Discord bot token

Find more skills at

clawhub.com

Warning Signs

Symptoms of a Neglected Agent

If your agent shows these symptoms, it needs care.

😶

Generic responses

Treatment: Update SOUL.md with stronger personality guidelines. Give examples of good vs. bad responses.

🔄

Repeating old context

Treatment: Clean MEMORY.md. Remove outdated entries. Add current projects and priorities.

🤷

Asking questions it should know the answer to

Treatment: Check if key info is in USER.md or MEMORY.md. Your agent can only remember what's written down.

🏃

Taking actions without asking

Treatment: Tighten AGENTS.md safety rules. Add explicit "ask before" list. Review security boundaries.

😴

Never proactive, only reactive

Treatment: Configure heartbeats. Give it standing tasks. Agents that only respond to commands never reach their potential.

🎭

Hallucinating facts or capabilities

Treatment: Add web search skill. Correct it firmly when wrong. Update SOUL.md: "When uncertain, say so."

Keep this protocol handy. Print it, bookmark it, or save it to your workspace.

OpenClaw Care Center