openclaw - 💡(How to fix) Fix Feature request: Shared global bootstrap prompt for all agents [1 participants]

Problem

When managing multiple agents in OpenClaw, there's no way to apply a shared set of rules/instructions to all agents from a single source of truth.

Every time a rule needs to change — say, a new operational policy, a security guideline, or an updated workflow — we have to manually update the AGENTS.md (or equivalent) for every single agent. This doesn't scale.

Proposed solution

Introduce a shared global bootstrap prompt mechanism. Something like a GLOBAL_BOOTSTRAP.md (or similar) that lives in a central location (e.g., ~/.openclaw/global-bootstrap.md) and is automatically injected as an incremental addition to every agent's initial context on session startup.

This would function like a company-wide employee handbook or operating manual:

• Maintain one file
• All agents get the updated content on their next session start
• No per-agent manual updates needed

Use case

• Security policies that apply to all agents (e.g., "never delete production data without approval")
• Operational procedures (e.g., "report incidents to the on-call engineer")
• Shared naming conventions, credential handling rules
• Any rule that should be universal across the entire agent fleet

Suggested implementation

1. A central file at ~/.openclaw/global-bootstrap.md
2. A config option like agents.defaults.globalBootstrapPath to point to it
3. On every agent session start, this file's content is injected after the agent's own workspace bootstrap files
4. Only one instance of the file needs to exist and be maintained

Alternatives considered

• Skills: Skills only inject metadata into the system prompt, not fixed content on every session. Not suitable for this use case.
• BOOTSTRAP.md: Only runs once on first workspace creation, not on every session.
• Shared skills directory: Still requires per-agent skill loading; doesn't guarantee injection on every session start.

Prior art

Many multi-agent frameworks (e.g., CrewAI, AutoGen, Microsoft AutoAgent) have some concept of a "shared system prompt" or "global instructions" that all agents inherit. OpenClaw is well-positioned to offer this given its existing multi-agent architecture.

Would love to hear if this aligns with the project's vision, and happy to contribute if someone points me in the right direction on implementation.