Summary
After upgrading to OpenClaw 2026.3.24 on macOS, multiple tools stopped receiving their required parameters and failed validation.
Environment
- OpenClaw:
2026.3.24
- OS: macOS Darwin 25.4.0 (arm64)
- Session/channel: Telegram direct chat
- Models tested:
kimi/k2p5
openrouter/google/gemini-2.5-pro-exp-03-25:free
- later session showed
openai-codex/gpt-5.4
Symptoms
Several tools received empty payloads like {} even though the session/model was attempting to call them with arguments.
Examples:
exec failed because command was missing
read failed because path / file_path was missing
gateway failed because action was missing
Representative errors:
Validation failed for tool "exec": command: must have required property 'command'
Missing required parameter: path
Validation failed for tool "gateway": action: must have required property 'action'
What still worked
Not all tools were broken:
session_status worked
- model switching worked
- after rollback,
agents_list also worked
This suggests a partial regression in tool argument passing/binding rather than a total runtime failure.
Reproduction
- Upgrade/install
OpenClaw 2026.3.24
- Start a normal session
- Attempt tool calls requiring parameters, such as:
exec(command="pwd")
read(path="SOUL.md")
gateway(action="restart")
- Observe that the tools receive
{} / missing required fields
Troubleshooting attempted
- started a new session
- restarted gateway
- reinstalled OpenClaw
- changed models
None of the above fixed the issue.
Workaround / Resolution
Rollback to 2026.3.23 fixed the problem.
After rollback, smoke test passed:
exec ✅
read ✅
session_status ✅
agents_list ✅
Likely cause
Regression introduced in 2026.3.24 affecting parameter passing for some tool bindings/runtime paths.
Summary
After upgrading to OpenClaw 2026.3.24 on macOS, multiple tools stopped receiving their required parameters and failed validation.
Environment
2026.3.24kimi/k2p5openrouter/google/gemini-2.5-pro-exp-03-25:freeopenai-codex/gpt-5.4Symptoms
Several tools received empty payloads like
{}even though the session/model was attempting to call them with arguments.Examples:
execfailed becausecommandwas missingreadfailed becausepath/file_pathwas missinggatewayfailed becauseactionwas missingRepresentative errors:
Validation failed for tool "exec": command: must have required property 'command'Missing required parameter: pathValidation failed for tool "gateway": action: must have required property 'action'What still worked
Not all tools were broken:
session_statusworkedagents_listalso workedThis suggests a partial regression in tool argument passing/binding rather than a total runtime failure.
Reproduction
OpenClaw 2026.3.24exec(command="pwd")read(path="SOUL.md")gateway(action="restart"){}/ missing required fieldsTroubleshooting attempted
None of the above fixed the issue.
Workaround / Resolution
Rollback to
2026.3.23fixed the problem.After rollback, smoke test passed:
exec✅read✅session_status✅agents_list✅Likely cause
Regression introduced in
2026.3.24affecting parameter passing for some tool bindings/runtime paths.