Bug type
Regression (worked before, now fails)
Beta release blocker
Yes
Summary
Description:
After updating to 2026.3.28 (f9b1079), agents cannot execute any commands. The exec tool fails validation with:
Validation failed for tool "exec": command: must have required property 'command'
Received arguments: {}Environment:
• OpenClaw version: 2026.3.28 (f9b1079)
• OS: macOS 15.3.2 (Darwin 25.3.0 arm64)
• Install method: npm global
• Model: kimi-coding/k2p5
Config (tools section):
"tools": {
"profile": "full",
"allow": ["*"],
"exec": {
"host": "gateway",
"security": "full",
"ask": "off"
}
}Tried:
• ✅ Latest version 2026.3.28
• ✅ Downgrade to 2026.2.21
• ✅ Config fix (exec settings with ask: "off")
• ✅ Multiple gateway restarts
• ✅ All "workarounds" from #39691
Result: Still broken. The agent generates commands but they never execute.
Impact: Completely blocks autonomous agent functionality.
Steps to reproduce
- Install/update OpenClaw:
npm install -g [email protected]. Verify config has exec enabled:
openclaw config get tools.profile # returns "full"
openclaw config get tools.exec # returns {"host":"gateway","security":"full","ask":"off"}3. Start/restart gateway:
openclaw gateway restart4. Ask agent to run any command:
• "List files in current directory"
• "Check what version of node is installed"
• "Run: echo 'hello'"
- Observe failure:
• Agent generates the command
• Validation error: command: must have required property 'command'
• Command never executes
• Agent falls back to asking user to run it manually
Expected: Command executes automatically
Actual: Validation fails, command displayed but not run
Expected behavior
When an agent requests to execute a command via the exec tool, the gateway should:
- Accept the command parameters
- Execute the command on the host system
- Return the output to the agent
Example of working flow (pre-2026.3.7):
User: "List files in current directory"
Agent: [calls exec with command="ls -la"]
System: [executes ls -la]
Agent: "Here are the files: [output]"Actual behavior (broken):
User: "List files in current directory"
Agent: [calls exec with command="ls -la"]
System: Validation failed for tool "exec": command: must have required property 'command'
Received arguments: {}
Agent: "I can't run commands directly. Please run: ls -la"The agent should be able to execute commands autonomously when tools.profile is set to "full" or "coding" and tools.exec.ask is "off".
Actual behavior
When an agent requests to execute a command via the exec tool, the gateway should:
- Accept the command parameters
- Execute the command on the host system
- Return the output to the agent
Example of working flow (pre-2026.3.7):
User: "List files in current directory"
Agent: [calls exec with command="ls -la"]
System: [executes ls -la]
Agent: "Here are the files: [output]"Actual behavior (broken):
User: "List files in current directory"
Agent: [calls exec with command="ls -la"]
System: Validation failed for tool "exec": command: must have required property 'command'
Received arguments: {}
Agent: "I can't run commands directly. Please run: ls -la"The agent should be able to execute commands autonomously when tools.profile is set to "full" or "coding" and tools.exec.ask is "off".
OpenClaw version
When an agent requests to execute a command via the exec tool, the gateway should: 1. Accept the command parameters 2. Execute the command on the host system 3. Return the output to the agent Example of working flow (pre-2026.3.7): User: "List files in current directory" Agent: [calls exec with command="ls -la"] System: [executes ls -la] Agent: "Here are the files: [output]"Actual behavior (broken): User: "List files in current directory" Agent: [calls exec with command="ls -la"] System: Validation failed for tool "exec": command: must have required property 'command' Received arguments: {} Agent: "I can't run commands directly. Please run: ls -la"The agent should be able to execute commands autonomously when tools.profile is set to "full" or "coding" and tools.exec.ask is "off".
Operating system
Operating system: macOS Sequoia 15.3.2 (Darwin 25.3.0, arm64 — Apple Silicon) Node version: v22.22.1 Shell: zsh
Install method
Install method: npm global npm install -g [email protected]
Model
Model: kimi-coding/k2p5 (primary) Also tested with fallback: • ollama/llama3.2 (configured but not reached due to exec failure) Model configuration: "models": { "providers": { "kimi-coding": { "baseUrl": "https://api.kimi.com/coding/", "api": "anthropic-messages" } } }
Provider / routing chain
Provider: Kimi Coding (kimi-coding) Provider ID: kimi-coding/k2p5 Base URL: https://api.kimi.com/coding/
Additional provider/model setup details
No response
Logs, screenshots, and evidence
Tool validation error (consistent):
Validation failed for tool "exec":
- command: must have required property 'command'
Received arguments:
{}Config verification:
$ cat ~/.openclaw/openclaw.json | grep -A5 '"tools"'
"tools": {
"profile": "full",
"allow": ["*"],
"exec": {
"host": "gateway",
"security": "full",
"ask": "off"
}
}Failed downgrade attempt:
npm error code EACCES
npm error syscall rename
npm error path /opt/homebrew/lib/node_modules/openclaw
...
npm error The operation was rejected by your operating system.Session crash during restart:
node:events:497
throw er; // Unhandled 'error' event
^
Error: write EPIPENo gateway logs — validation fails before command reaches gateway.Impact and severity
Impact: Critical — Core functionality completely broken
• 100% of exec tool calls fail
• Agents cannot execute any commands autonomously
• All file operations, system commands, package installs blocked
• User must manually run every command agent generates
• Defeats purpose of autonomous agent
Security considerations:
• Bug forces manual command execution (safer, but inconvenient)
• No security vulnerability — just broken functionality
• No privilege escalation risk
• No data exposure risk
Severity: P0/Critical — Blocks all autonomous operations
Additional information
• Consistency: Bug occurs 100% of the time — every exec call fails
• Error details: Validation shows Received arguments: {} — the command parameter is not being passed through
• Gateway logs: No exec attempt appears in gateway logs — validation fails before reaching gateway
• Related issues: Persists despite fixes from #39691 — config has tools.profile: "full" and tools.exec.ask: "off" but still fails
• Other tools affected: write tool also intermittently fails with content: must have required property 'content'
• Session stability: Attempting to downgrade OpenClaw version caused session crashes
• Workaround: User must manually run commands that agent generates
Suspected cause: Tool parameter validation layer is not receiving arguments from model, or model is not sending parameters in expected format.
Bug type
Regression (worked before, now fails)
Beta release blocker
Yes
Summary
Description:
After updating to 2026.3.28 (f9b1079), agents cannot execute any commands. The exec tool fails validation with:
Validation failed for tool "exec": command: must have required property 'command'
Received arguments: {}Environment:
• OpenClaw version: 2026.3.28 (f9b1079)
• OS: macOS 15.3.2 (Darwin 25.3.0 arm64)
• Install method: npm global
• Model: kimi-coding/k2p5
Config (tools section):
"tools": {
"profile": "full",
"allow": ["*"],
"exec": {
"host": "gateway",
"security": "full",
"ask": "off"
}
}Tried:
• ✅ Latest version 2026.3.28
• ✅ Downgrade to 2026.2.21
• ✅ Config fix (exec settings with ask: "off")
• ✅ Multiple gateway restarts
• ✅ All "workarounds" from #39691
Result: Still broken. The agent generates commands but they never execute.
Impact: Completely blocks autonomous agent functionality.
Steps to reproduce
npm install -g [email protected]. Verify config has exec enabled:
openclaw config get tools.profile # returns "full"
openclaw config get tools.exec # returns {"host":"gateway","security":"full","ask":"off"}3. Start/restart gateway:
openclaw gateway restart4. Ask agent to run any command:
• "List files in current directory"
• "Check what version of node is installed"
• "Run: echo 'hello'"
• Agent generates the command
• Validation error: command: must have required property 'command'
• Command never executes
• Agent falls back to asking user to run it manually
Expected: Command executes automatically
Actual: Validation fails, command displayed but not run
Expected behavior
When an agent requests to execute a command via the exec tool, the gateway should:
Example of working flow (pre-2026.3.7):
User: "List files in current directory"
Agent: [calls exec with command="ls -la"]
System: [executes ls -la]
Agent: "Here are the files: [output]"Actual behavior (broken):
User: "List files in current directory"
Agent: [calls exec with command="ls -la"]
System: Validation failed for tool "exec": command: must have required property 'command'
Received arguments: {}
Agent: "I can't run commands directly. Please run: ls -la"The agent should be able to execute commands autonomously when tools.profile is set to "full" or "coding" and tools.exec.ask is "off".
Actual behavior
When an agent requests to execute a command via the exec tool, the gateway should:
Example of working flow (pre-2026.3.7):
User: "List files in current directory"
Agent: [calls exec with command="ls -la"]
System: [executes ls -la]
Agent: "Here are the files: [output]"Actual behavior (broken):
User: "List files in current directory"
Agent: [calls exec with command="ls -la"]
System: Validation failed for tool "exec": command: must have required property 'command'
Received arguments: {}
Agent: "I can't run commands directly. Please run: ls -la"The agent should be able to execute commands autonomously when tools.profile is set to "full" or "coding" and tools.exec.ask is "off".
OpenClaw version
When an agent requests to execute a command via the exec tool, the gateway should: 1. Accept the command parameters 2. Execute the command on the host system 3. Return the output to the agent Example of working flow (pre-2026.3.7): User: "List files in current directory" Agent: [calls exec with command="ls -la"] System: [executes ls -la] Agent: "Here are the files: [output]"Actual behavior (broken): User: "List files in current directory" Agent: [calls exec with command="ls -la"] System: Validation failed for tool "exec": command: must have required property 'command' Received arguments: {} Agent: "I can't run commands directly. Please run: ls -la"The agent should be able to execute commands autonomously when tools.profile is set to "full" or "coding" and tools.exec.ask is "off".
Operating system
Operating system: macOS Sequoia 15.3.2 (Darwin 25.3.0, arm64 — Apple Silicon) Node version: v22.22.1 Shell: zsh
Install method
Install method: npm global npm install -g [email protected]
Model
Model: kimi-coding/k2p5 (primary) Also tested with fallback: • ollama/llama3.2 (configured but not reached due to exec failure) Model configuration: "models": { "providers": { "kimi-coding": { "baseUrl": "https://api.kimi.com/coding/", "api": "anthropic-messages" } } }
Provider / routing chain
Provider: Kimi Coding (kimi-coding) Provider ID: kimi-coding/k2p5 Base URL: https://api.kimi.com/coding/
Additional provider/model setup details
No response
Logs, screenshots, and evidence
Impact and severity
Impact: Critical — Core functionality completely broken
• 100% of exec tool calls fail
• Agents cannot execute any commands autonomously
• All file operations, system commands, package installs blocked
• User must manually run every command agent generates
• Defeats purpose of autonomous agent
Security considerations:
• Bug forces manual command execution (safer, but inconvenient)
• No security vulnerability — just broken functionality
• No privilege escalation risk
• No data exposure risk
Severity: P0/Critical — Blocks all autonomous operations
Additional information
• Consistency: Bug occurs 100% of the time — every exec call fails
• Error details: Validation shows Received arguments: {} — the command parameter is not being passed through
• Gateway logs: No exec attempt appears in gateway logs — validation fails before reaching gateway
• Related issues: Persists despite fixes from #39691 — config has tools.profile: "full" and tools.exec.ask: "off" but still fails
• Other tools affected: write tool also intermittently fails with content: must have required property 'content'
• Session stability: Attempting to downgrade OpenClaw version caused session crashes
• Workaround: User must manually run commands that agent generates
Suspected cause: Tool parameter validation layer is not receiving arguments from model, or model is not sending parameters in expected format.