mystic kettle
Hi everyone!
Let me describe my situation — maybe it's just a mental issue 😄
I have many side projects on my GitHub account, and I’d like to manage them using OpenClaw + OpenCode.
Goal
My idea is to have one Telegram topic per project.
So, by writing in a specific topic, the system understands which project I'm referring to.
Current Setup
In OpenClaw (if I understood correctly), each agent owns its own workspace.
For example, I might have:
- a PR Reviewer agent
- a Software Engineer agent
- a Triage/Issue agent
Each of them works in its own workspace and maintains its own memory.md.
The Problem
Across all my projects, I would like to reuse the same agents.
But here's the issue:
- If agents clone and work on Project A, their workspace and memory get populated with Project A context
- Then, if I use the same agents on Project B, their memory could get "contaminated" with context from Project A
This breaks the isolation between projects, which is something I’d really like to preserve.
Questions
- Am I misunderstanding how OpenClaw workspaces and agents are supposed to be used?
- How do you handle multi-project setups with shared agent roles?
- Is there a recommended pattern for isolating context per project (without duplicating agents N × M times)?
What I'm Thinking
Maybe I should rethink the model (e.g., agents per project instead of per role), but that feels a bit redundant.
Would love to hear how others are solving this!
grave bolt