openclaw - 💡(How to fix) Fix [Bug]: Session not recovered after gateway crash — orphaned tool_use blocks cause permanent LLM rejection [1 participants]

openclaw2026-03-24 16:03:51

ON THIS PAGE

Recommended Tools

×6

Utilities matched from this issue’s tags and category — try them while you read without losing context.

GitHub issue graph ai analysis

Paste a GitHub issue URL. We fetch that issue, discover linked issues from bodies/comments/timeline, collect linked pull requests, and produce a structured English report.

The report is written in English Markdown for sharing and archival.

GitHub issue URL

Helpful · Quick feedback

GitHub stats

openclaw/openclaw#53781•Fetched 2026-04-08 01:23:29

View on GitHub

Comments

Participants

Timeline

Reactions

Author

shiftymix

Participants

shiftymix

Timeline (top)

cross-referenced ×1

When the gateway crashes mid-tool-execution, the session JSONL contains a tool_use block with no corresponding tool_result. On restart, every subsequent LLM request is permanently rejected by the Anthropic API:

LLM request rejected: messages.91: tool_use ids were found without tool_result blocks
immediately after: functionswrite0374f29f4. Each tool_use block must have a corresponding
tool_result block in the next message.

The agent is stuck in an unrecoverable loop — it cannot read its own history, cannot patch the session, and cannot self-heal. Manual intervention (session clear + re-brief) is required every time this happens.

Error Message

Any message to the agent returns the tool_use ids were found without tool_result error indefinitely On session load, OpenClaw should scan the last N messages for unpaired tool_use blocks and inject a synthetic tool_result with an error payload:
If any are unpaired, inject a synthetic error tool_result immediately after the orphaned tool_use

Root Cause

Gateway crash (e.g. Discord WS 1006 — see #53644) kills the Node process between:

Writing the tool_use block to the session JSONL
Executing the tool and writing the tool_result

On reload, the session history is structurally invalid per Anthropic's API contract.

Fix Action

Fix / Workaround

Agent becomes permanently unresponsive after any mid-tool gateway crash
Session context is lost (users must clear and re-brief the agent)
No user-configurable workaround exists today
Related to #53644 (gateway crash on Discord WS 1006) — that issue causes the crash; this issue describes the downstream consequence

Code Example

LLM request rejected: messages.91: tool_use ids were found without tool_result blocks
immediately after: functionswrite0374f29f4. Each tool_use block must have a corresponding
tool_result block in the next message.

---

{
  "type": "tool_result",
  "tool_use_id": "<orphaned_id>",
  "content": "[Gateway crash: tool execution did not complete. Please retry.]",
  "is_error": true
}

RAW_BUFFERClick to expand / collapse

Summary

LLM request rejected: messages.91: tool_use ids were found without tool_result blocks
immediately after: functionswrite0374f29f4. Each tool_use block must have a corresponding
tool_result block in the next message.

Root Cause

Gateway crash (e.g. Discord WS 1006 — see #53644) kills the Node process between:

Writing the tool_use block to the session JSONL
Executing the tool and writing the tool_result

On reload, the session history is structurally invalid per Anthropic's API contract.

Reproduction Steps

Agent invokes a tool (e.g. write, exec)
Gateway process crashes or is killed before the tool result is recorded
Gateway restarts, session reloads
Any message to the agent returns the tool_use ids were found without tool_result error indefinitely

Expected Behavior

On session load, OpenClaw should scan the last N messages for unpaired tool_use blocks and inject a synthetic tool_result with an error payload:

{
  "type": "tool_result",
  "tool_use_id": "<orphaned_id>",
  "content": "[Gateway crash: tool execution did not complete. Please retry.]",
  "is_error": true
}

This restores API validity and lets the agent acknowledge the failure and continue.

Suggested Fix

In the session hydration / history-replay path, before sending messages to the LLM provider:

Walk the message array and build a set of seen tool_use IDs
For each tool_use ID, verify a tool_result follows in the next message
If any are unpaired, inject a synthetic error tool_result immediately after the orphaned tool_use
Log a warning so the user knows a recovery injection occurred

This is a pure defensive measure — the agent recovers gracefully and can retry the failed tool call.

Impact

Agent becomes permanently unresponsive after any mid-tool gateway crash
Session context is lost (users must clear and re-brief the agent)
No user-configurable workaround exists today
Related to #53644 (gateway crash on Discord WS 1006) — that issue causes the crash; this issue describes the downstream consequence

Environment

OpenClaw v2026.3.12
Node.js v24.13.1
Windows 10 x64
Provider: anthropic (claude-sonnet-4-6)

extent analysis

Fix Plan

To resolve the issue, we need to implement a session hydration fix that injects synthetic tool_result blocks for unpaired tool_use IDs. Here are the steps:

Modify the session hydration code to walk the message array and build a set of seen tool_use IDs.
For each tool_use ID, verify a tool_result follows in the next message.
If any are unpaired, inject a synthetic error tool_result immediately after the orphaned tool_use.
Log a warning to notify the user of the recovery injection.

Example code snippet:

const hydrateSession = (messages) => {
  const toolUseIds = new Set();
  const syntheticResults = [];

  messages.forEach((message) => {
    if (message.type === 'tool_use') {
      toolUseIds.add(message.tool_use_id);
    } else if (message.type === 'tool_result') {
      toolUseIds.delete(message.tool_use_id);
    }
  });

  toolUseIds.forEach((id) => {
    const syntheticResult = {
      type: 'tool_result',
      tool_use_id: id,
      content: '[Gateway crash: tool execution did not complete. Please retry.]',
      is_error: true,
    };
    syntheticResults.push(syntheticResult);
  });

  // Inject synthetic results into the message array
  const updatedMessages = messages.map((message) => {
    if (message.type === 'tool_use' && toolUseIds.has(message.tool_use_id)) {
      return [message, syntheticResults.find((result) => result.tool_use_id === message.tool_use_id)];
    }
    return message;
  });

  return updatedMessages;
};

Verification

To verify the fix, restart the gateway and send a message to the agent. The agent should now be able to recover from the unpaired tool_use block and continue processing messages.

Extra Tips

Make sure to log a warning when injecting synthetic results to notify the user of the recovery injection.
Consider implementing a retry mechanism for failed tool calls to improve user experience.
Review the gateway crash issue (#53644) and implement a fix to prevent crashes from occurring in the first place.

Vote matrix · Quick signals

Works

Did the solution work? Tap to confirm.

Easy Fix

Was it a quick fix?

Time Saver

Did it save you time?

Blocking

Was it severely blocking?

Common Issue

Are others likely hitting this too?

Flaky / Intermittent

Is it intermittent?

Verified / Reproducible

Can you reproduce it reliably?

#api #ssr #installation #tensor shape #request timeout

Still need to ship something?

×6

Another batch ranked right after the header list — different links, same matching logic.

Data

Security

Network

Code

UI/UX

Text

System

Multimedia

Protocol

API

Engineering

openclaw - 💡(How to fix) Fix [Bug]: Session not recovered after gateway crash — orphaned tool_use blocks cause permanent LLM rejection [1 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

Root Cause

Fix Action

Fix / Workaround

Code Example

Summary

Root Cause

Reproduction Steps

Expected Behavior

Suggested Fix

Impact

Environment

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

TRENDING

openclaw - 💡(How to fix) Fix [Bug]: Session not recovered after gateway crash — orphaned tool_use blocks cause permanent LLM rejection [1 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

Root Cause

Fix Action

Fix / Workaround

Code Example

Summary

Root Cause

Reproduction Steps

Expected Behavior

Suggested Fix

Impact

Environment

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

RELATED_DISCOVERY

TRENDING