Support per-agent containerTag for multi-agent isolation

[kn0wn]

Problem

When running OpenClaw with multiple agents (via agents.list + bindings), all agents currently share the same Supermemory containerTag. This creates a memory pollution risk where one agent's stored memories become visible to another agent's recall context.

Use Case

We have two agents:

• main — Claude Opus, handles personal/private conversations
• employee — Gemini Flash, handles a shared Slack channel with team members

Both agents share the same Supermemory API key and default containerTag. If someone in Slack tells the employee agent to "remember X", that memory can later surface in the main agent's context — even though the sessions are meant to be isolated.

Proposed Solution

Allow containerTag to be templated with agent context, similar to how workspaces work per-agent.

Option A: Template variable in config

{
  "plugins": {
    "entries": {
      "openclaw-supermemory": {
        "config": {
          "containerTag": "openclaw_{hostname}_{agentId}"
        }
      }
    }
  }
}

Option B: Automatic per-agent namespacing
If multi-agent is detected, automatically append _{agentId} to the container tag.

Option C: Per-agent plugin config (requires OpenClaw core support)
Allow overriding plugin config at the agent level in agents.list[].plugins.

Current Workaround

None that preserve functionality. Options are:

1. Accept shared memory (security/privacy risk)
2. Disable supermemory for secondary agents (lose the feature)
3. Use completely separate Supermemory accounts per agent (cost/management overhead)

Environment

• OpenClaw: 2026.2.1
• openclaw-supermemory: 1.0.3
• Multi-agent config with bindings

[Dhravya]

This is a great idea, working on merging this asap