openclaw - 💡(How to fix) Fix [Feature Request] Agent loses identity context when switching between messaging channels [1 participants]

Problem

When an OpenClaw agent is configured with a custom identity (name, personality, role description), this identity context is not preserved across different messaging channels (e.g., webchat vs. Feishu vs. Discord).

Each channel creates a separate session, and the agent starts as if it has no prior identity knowledge — even though identity files are configured in the workspace.

Specific Example

1. User defines a custom AI assistant persona via workspace files (SOUL.md, IDENTITY.md) in the webchat session
2. The same agent is messaged via a different channel (e.g., Feishu)
3. The Feishu session behaves as if it has no memory of the configured identity — it doesn't know its own name, role, or communication style

Expected Behavior

When an agent has a configured identity, that identity should be consistent regardless of which channel the user communicates through. Different channels should function as different "contact entry points" to the same agent — not as separate agent instances.

Technical Context

Currently, semantic memory search can find fragments of identity information, but:

• It relies on passive keyword matching, not structured identity injection
• SOUL.md and IDENTITY.md are not automatically loaded on session start for all channels
• Each session bootstrap appears to start from a blank context for identity-related files

Suggested Solutions

• Option A: On session initialization, automatically inject the full identity file contents (SOUL.md, IDENTITY.md) as system-level context, before any user message is processed
• Option B: Provide a shared "global identity layer" that all channel sessions reference, so identity is loaded once and reused across sessions
• Option C: Add a configuration flag like agent.identity.autoInject: true that ensures identity files are always the first thing loaded during bootstrap on any channel