openclaw - 💡(How to fix) Fix [Bug]:Agent fabricates tool output for factual queries — calendar tool never invoked despite instrumentation [1 participants]

Official PRs (…)
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.

Helpful · Quick feedback

Loading…
GitHub stats
openclaw/openclaw#63289Fetched 2026-04-09 07:55:48
View on GitHub
Comments
0
Participants
1
Timeline
2
Reactions
0
Participants
Timeline (top)
labeled ×2

I am a non-developer end user running OpenClaw 2026.4.1 in a Lume VM, and that this was discovered when the agent gave me wrong information about my own on-call schedule. I tried to debug with the help of Opus. Opus determined that OpenClaw was fabricating responses. I've attached a file which I had Opus create outlining it's efforts to debug.

Root Cause

I am a non-developer end user running OpenClaw 2026.4.1 in a Lume VM, and that this was discovered when the agent gave me wrong information about my own on-call schedule. I tried to debug with the help of Opus. Opus determined that OpenClaw was fabricating responses. I've attached a file which I had Opus create outlining it's efforts to debug.

RAW_BUFFERClick to expand / collapse

Bug type

Behavior bug (incorrect output/state without crash)

Beta release blocker

No

Summary

I am a non-developer end user running OpenClaw 2026.4.1 in a Lume VM, and that this was discovered when the agent gave me wrong information about my own on-call schedule. I tried to debug with the help of Opus. Opus determined that OpenClaw was fabricating responses. I've attached a file which I had Opus create outlining it's efforts to debug.

Steps to reproduce

Please see file uploaded.

Expected behavior

When the agent receives a factual, time-sensitive question (e.g. "What's on my calendar today?"), it should invoke the authoritative tool for that domain — in this case gog calendar events — and return the result from that live call. Every such query should produce a fresh tool invocation, never an answer reconstructed from prior context or from memory of earlier tool outputs. Any reply the agent produces should leave a corresponding trace in the gateway log or tool instrumentation; if the trace is absent, the agent should not have produced an answer.

Actual behavior

Here's a clean version for the "Actual behaviour" field:

The agent returns a confident, plausible-sounding answer without invoking any calendar tool. Instrumentation on the gog binary (a shim wrapper at /opt/homebrew/bin/gog that logs every invocation to /tmp/gog-shim.log) confirms the binary is never called during these replies, and no tool call appears in ~/.openclaw/logs/gateway.log between receiving the user message and sending the reply. The replies are either hallucinated or reconstructed from content the agent saw earlier in its context window. On 2026-04-08 the agent told the user "Nothing scheduled. You're clear all day" when ground truth (verified by running gog directly) was 6 Call Calendar events and 4 Imaging Reading Schedule events. Even an explicit instruction — "Run jarvis-cal today in the shell and paste the raw output, nothing else" — produced a reply with correctly-formatted markdown output, yet /tmp/gog-shim.log remained empty, confirming the tool was never invoked and the output was fabricated from prior context. Opus 4.6

OpenClaw version

OpenClaw 2026.4.1.

Operating system

macOS Tahoe, ver. 25.4

Install method

Installed via npm (openclaw package), global install at /opt/homebrew/lib/node_modules/openclaw, launched by launchd under label ai.openclaw.gateway

Model

claude-haiku-4-5-20251001),

Provider / routing chain

Local gateway → Anthropic API direct

Additional provider/model setup details

No response

Logs, screenshots, and evidence

Impact and severity

No response

Additional information

No response

extent analysis

TL;DR

The issue can likely be resolved by ensuring that OpenClaw properly invokes the authoritative tool for domain-specific queries, such as the gog calendar events tool, instead of fabricating responses from prior context.

Guidance

  • Verify that the gog calendar events tool is correctly configured and accessible to OpenClaw, and that the shim wrapper at /opt/homebrew/bin/gog is functioning as expected.
  • Check the OpenClaw configuration to ensure that it is set up to invoke the gog tool for calendar-related queries, rather than relying on cached or reconstructed responses.
  • Review the gateway log and tool instrumentation to confirm that the gog tool is being invoked correctly and that the expected traces are present.
  • Consider updating OpenClaw to a newer version or modifying the configuration to prioritize live tool invocations over cached responses.

Example

No specific code snippet is provided, as the issue appears to be related to configuration and tool invocation rather than code-level errors.

Notes

The issue may be specific to the OpenClaw version (2026.4.1) or the claude-haiku-4-5-20251001 model, and further investigation may be needed to determine the root cause. Additionally, the lack of information on the Impact and severity and Additional information sections may limit the ability to provide a comprehensive solution.

Recommendation

Apply workaround: Modify the OpenClaw configuration to prioritize live tool invocations over cached responses, and verify that the gog tool is correctly configured and accessible. This should help ensure that OpenClaw provides accurate and up-to-date information for domain-specific queries.

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?
Loading…

FAQ

Expected behavior

When the agent receives a factual, time-sensitive question (e.g. "What's on my calendar today?"), it should invoke the authoritative tool for that domain — in this case gog calendar events — and return the result from that live call. Every such query should produce a fresh tool invocation, never an answer reconstructed from prior context or from memory of earlier tool outputs. Any reply the agent produces should leave a corresponding trace in the gateway log or tool instrumentation; if the trace is absent, the agent should not have produced an answer.

Still need to ship something?

×6

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

Back to top recommendations

TRENDING