Summary

When a tool fails transiently (e.g., an edit with mismatched content) but the agent retries and succeeds, the tool error warning is still delivered to the user's channel as a visible message:

⚠️ 📝 Edit: in ~/IDENTITY.md (521 chars) failed

This creates a confusing UX where the user sees:

1. "Job done!" (success message from agent)
2. "⚠️ Edit failed" (transient error bubble)

The agent self-corrected and succeeded, but the user sees an error.

Current behavior

• Tool failure summaries are always visible in normal mode (per docs: "Tool failure summaries remain visible in normal mode")
• suppressToolErrorWarnings exists but only applies to heartbeat runs
• /verbose off only hides detail suffixes, not the warning itself

Requested feature

A config option to suppress tool error warnings when the agent's final reply indicates success:

{
  agents: {
    defaults: {
      toolErrorDelivery: "final", // "all" (default) | "final" | "off"
    },
  },
}

• all: current behavior (show all tool errors immediately)
• final: only show tool errors if the final reply doesn't indicate success
• off: never show tool error warnings to users

Alternatively, extend suppressToolErrorWarnings to apply globally (not just heartbeats).

Use case

Hosted/managed OpenClaw deployments where non-technical users are confused by transient errors that the agent successfully recovered from.