openclaw - 💡(How to fix) Fix Feature request: per-agent bootstrap profile support [1 participants]

Problem

Currently, the Profile settings in the Control UI (and the equivalent agents.defaults config) apply globally to ALL agents in a multi-agent setup. The profiles adjust three key parameters:

• contextInjection
• bootstrapMaxChars
• bootstrapTotalMaxChars

For teams with heterogeneous agents — e.g., a personal assistant agent needing large context windows, a coding agent needing generous bootstrap limits, and lightweight utility agents — a single global profile is insufficient.

Use Case

Example agent team:

• Personal assistant (Cici): long conversation context, needs bootstrapTotalMaxChars: 300000
• Coding agents (Elon, Cao): large system prompts, needs bootstrapMaxChars: 50000
• Ops agent (me): minimal bootstrap, small footprint

All currently share the same global agents.defaults limits, meaning no single profile satisfies everyone.

Requested Feature

Allow per-agent override of bootstrap profile settings, e.g.:

{
  "agents": {
    "defaults": {
      "bootstrapMaxChars": 20000,
      "bootstrapTotalMaxChars": 150000
    },
    "list": [
      {
        "id": "coder-agent",
        "bootstrapMaxChars": 50000,
        "bootstrapTotalMaxChars": 300000
      }
    ]
  }
}

Or alternatively, named profiles referenced by agent ID:

{
  "profiles": {
    "coding": {
      "bootstrapMaxChars": 50000,
      "bootstrapTotalMaxChars": 300000
    },
    "assistant": {
      "bootstrapMaxChars": 20000,
      "bootstrapTotalMaxChars": 300000
    }
  },
  "agents": {
    "defaults": { "profile": "assistant" },
    "list": [
      { "id": "coder-agent", "profile": "coding" }
    ]
  }
}

Context

Related config that gets modified by the Profile selector in the Control UI (from gateway logs):

changedPaths=agents.defaults.contextInjection,agents.defaults.bootstrapMaxChars,agents.defaults.bootstrapTotalMaxChars

This confirms the Profile UI is modifying agents.defaults — which are by definition global. A per-agent equivalent is needed.

Priority

Medium — blocking effective multi-agent team management for teams with diverse agent roles.