Skip to content

Agent silent failure: tool errors cause no response to user #12595

@bennyaph1211

Description

@bennyaph1211

Environment

  • OpenClaw version: 2026.2.6-3
  • OS: Linux (WSL2) 6.6.87.2-microsoft-standard-WSL2
  • Runtime: systemd
  • Model: google-antigravity/claude-sonnet-4-5-thinking
  • Channel: Telegram

Bug Summary

Agent experiences tool execution error and fails to send any response to the user, resulting in complete silence instead of graceful error handling. User is left without any indication that something went wrong.

Expected Behavior

When a tool execution fails:

  1. Agent should catch the error
  2. Send user a response explaining the issue
  3. Optionally retry or suggest alternatives

Actual Behavior

When certain tool errors occur:

  1. Tool fails with error (e.g., "missing tool result in session history")
  2. Agent sends NO response at all
  3. Session appears to "freeze" or "sleep"
  4. User must send another message to wake agent
  5. Agent then continues as if nothing happened

Reproduction Steps

  1. Execute a tool command that triggers a session history error
  2. In this case: exec command to create directory (mkdir -p)
  3. Tool returns: "[openclaw] missing tool result in session history; inserted synthetic error result for transcript repair."
  4. Observe: Agent fails to send any response
  5. Send another message to agent
  6. Observe: Agent wakes up and responds normally

Timeline Evidence

User perspective:

  • 17:51 - User: "Start with 3 at first. Create 1 log file per day about what you did"
  • 17:51-18:27 - SILENCE (36 minutes)
  • 18:27 - User: "Hi ?"
  • 18:28 - Agent: Responds and completes the task

What happened internally:

  • Agent attempted mkdir command
  • Hit "missing tool result in session history" error
  • Generated no user-facing response
  • Session effectively froze until next user message

Error Message

[openclaw] missing tool result in session history; inserted synthetic error result for transcript repair.

Impact

Critical UX issue:

  • User has no idea if agent is working, broken, or stuck
  • Creates confusion and breaks trust
  • Forces user to "ping" agent to check if it's alive
  • Degrades user experience significantly

Expected: Agent should ALWAYS respond, even if just to say "Error occurred, retrying..."

Possible Root Causes

  1. Tool error handling doesn't guarantee response generation
  2. Session history repair mechanism blocks response pipeline
  3. No fallback/catch-all response for unhandled errors
  4. Error might occur after tool call but before response generation

Suggested Fix

Add error handling layer that ensures:

  1. Every user message MUST receive a response
  2. Tool failures generate user-facing error messages
  3. "Synthetic error results" still trigger graceful degradation
  4. Timeout mechanism if no response generated after N seconds

Notes

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions