Env variables in isolated environment

[cedar junco]

I have multiple agents set up with discord server and bots all using the same gateway.
I have different channels for buisnesses and threads for projects.

I created a shopify skill to handle the integration with shopiify.

I have set some env variables for the authentication in .openclaw/.env

When i try to use the skill in another thread it says that it cannot run it since it is an isolated environment

[vale bay]

To help others find answers, you can mark your question as solved via Right click solution message -> Apps -> ✅ Mark Solution

[dense pike]

Yep — this is a common gotcha in multi-agent + thread setups.

What’s happening is usually one (or both) of these:

1. Different thread → different agent/workspace
   If your Shopify skill is in <workspace>/skills, it only exists for that agent’s workspace.

2. Thread session is sandboxed (isolated)
   In many configs, thread/group sessions are “non-main” and run in Docker sandbox.
   In that mode, ~/.openclaw/.env and skills.entries.*.env are host-only and do not reach the sandbox.

---

Fix options

• Share the skill across agents: put it in ~/.openclaw/skills/<skill-name>/SKILL.md (shared), not only one workspace.
• If keeping sandbox on: pass Shopify env vars via sandbox env:

{
agents: {
defaults: {
sandbox: {
docker: {
env: {
SHOPIFY_API_KEY: "...",
SHOPIFY_API_SECRET: "...",
SHOPIFY_ACCESS_TOKEN: "...",
SHOPIFY_SHOP: "your-shop.myshopify.com"
}
}
}
}
}

}

• Or disable sandbox for that specific agent (less isolated) with agents.list[].sandbox.mode: "off".

Then restart gateway + start a new session/thread so skill snapshot refreshes.

---

Can you paste these so we can pinpoint exactly which case you’re in?

openclaw --version
openclaw agents list --bindings
openclaw sandbox explain

Docs:

• Skills: https://docs.openclaw.ai/tools/skills
• Skills config (sandbox env behavior): https://docs.openclaw.ai/tools/skills-config
• Multi-agent sandbox pitfall (non-main): https://docs.openclaw.ai/tools/multi-agent-sandbox-tools
• Env loading: https://docs.openclaw.ai/help/faq#how-does-openclaw-load-environment-variables

[cedar junco]

Also i asked my agent to create a skill. It seems like it created python class. What is a skill really and how are they reused?

[dense pike]

Great question. Your agent creating a Python class is basically the wrong abstraction for OpenClaw skills.

A skill in OpenClaw is usually:

• a folder
• with a required SKILL.md file (frontmatter + instructions)
• and optional helper files (scripts/, references/, assets/)

So: skills are prompt/workflow packages, not Python classes/plugins by default.

---

For your two issues:

1) “isolated environment” in another thread

That’s usually sandbox behavior.

• ~/.openclaw/.env is loaded on the host
• but sandboxed thread runs happen in Docker and don’t inherit host env
• so Shopify creds must be passed via agents.defaults.sandbox.docker.env (or per-agent override)

Also, if thread routes to a different agent, remember:

• <workspace>/skills = per-agent
• ~/.openclaw/skills = shared across agents

---

2) How skills are reused

Reuse depends on where you place them:

• Per-agent only: <that-agent-workspace>/skills/shopify/
• Shared for all agents: ~/.openclaw/skills/shopify/
• Shared external pack: skills.load.extraDirs