openclaw - 💡(How to fix) Fix [Bug] @mariozechner/pi-ai tool validation throws on empty args instead of returning tool_error — regression from 2026.3.22 [2 comments, 1 participants]

openclaw2026-03-24 12:36:06

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#53671•Fetched 2026-04-08 01:25:03

View on GitHub

Comments

Participants

Timeline

Reactions

Author

lihan891127

Participants

lihan891127

Timeline (top)

commented ×2

Error Message

if (!validate(args)) { return { type: "tool_error", error: Validation failed: ... }; // instead of: throw new Error(...) }

Root Cause

2026.3.22 introduced @mariozechner/pi-ai as an external dependency. Its validateToolArguments() function uses AJV strict validation and throws on missing required fields:

// @mariozechner/pi-ai/dist/utils/validation.js (v0.61.1)
export function validateToolArguments(tool, toolCall) {
    const validate = ajv.compile(tool.parameters);
    const args = structuredClone(toolCall.arguments);
    if (validate(args)) return args;
    throw new Error(`Validation failed for tool "${toolCall.name}":\n...`);  // ← hard throw
}

In 2026.3.13, OpenClaw used its own bundled pi-ai with no AJV validation layer — empty args were passed through silently. The behavior change is a silent breaking change: models that already sent occasional empty tool calls now experience full turn failures instead of recoverable errors.

Fix Action

Temporary Workaround

Use tools.byProvider to deny problematic tools for affected providers:

{
  tools: {
    byProvider: {
      "kimi-coding": { deny: ["exec", "process"] },
      "kimi": { deny: ["exec", "process"] }
    }
  }
}

This prevents the validation from being reached, but removes exec functionality for these models entirely.

Code Example

// @mariozechner/pi-ai/dist/utils/validation.js (v0.61.1)
export function validateToolArguments(tool, toolCall) {
    const validate = ajv.compile(tool.parameters);
    const args = structuredClone(toolCall.arguments);
    if (validate(args)) return args;
    throw new Error(`Validation failed for tool "${toolCall.name}":\n...`);  // ← hard throw
}

---

Validation failed for tool "exec":
command: must have required property 'command'

Received arguments: {}

---

if (!validate(args)) {
    return { type: "tool_error", error: `Validation failed: ...` };
    // instead of: throw new Error(...)
}

---

{
  tools: {
    byProvider: {
      "kimi-coding": { deny: ["exec", "process"] },
      "kimi": { deny: ["exec", "process"] }
    }
  }
}

---

{
  models: {
    providers: {
      "kimi": {
        toolCallValidation: "warn"  // "strict" | "warn" | "off"
      }
    }
  }
}

RAW_BUFFERClick to expand / collapse

Problem Description

Since upgrading from 2026.3.13 → 2026.3.22, certain third-party models (confirmed: Moonshot AI kimi-latest via openai-completions, and kimi-coding/k2p5 via anthropic-messages) occasionally emit tool calls with empty arguments {} for the exec tool. This now causes a hard exception that terminates the entire agent turn, instead of being handled gracefully.

Root Cause

2026.3.22 introduced @mariozechner/pi-ai as an external dependency. Its validateToolArguments() function uses AJV strict validation and throws on missing required fields:

// @mariozechner/pi-ai/dist/utils/validation.js (v0.61.1)
export function validateToolArguments(tool, toolCall) {
    const validate = ajv.compile(tool.parameters);
    const args = structuredClone(toolCall.arguments);
    if (validate(args)) return args;
    throw new Error(`Validation failed for tool "${toolCall.name}":\n...`);  // ← hard throw
}

Environment

OpenClaw version: 2026.3.23-2 (regression introduced in 2026.3.22)
@mariozechner/pi-ai version: 0.61.1
Affected models:
- kimi/kimi-latest (provider api: openai-completions)
- kimi-coding/k2p5 (provider api: anthropic-messages)
OS: macOS 25.3.0 (darwin arm64)

Observed Behavior

Validation failed for tool "exec":
command: must have required property 'command'

Received arguments: {}

This error is thrown as an uncaught exception, terminating the agent turn entirely. High frequency — occurs on most exec tool call attempts by these models.

Expected Behavior

Validation failures on tool call arguments should be handled gracefully as a tool_error result returned to the model, not a hard exception. This allows the model to recover (e.g., retry with correct arguments) rather than crashing the turn.

The equivalent behavior would be:

if (!validate(args)) {
    return { type: "tool_error", error: `Validation failed: ...` };
    // instead of: throw new Error(...)
}

Why This Matters

In 2026.3.13: empty {} args from model → silently passed through → recoverable downstream error
In 2026.3.22+: empty {} args from model → AJV throws → entire turn crashes

This affects any third-party model (especially those using openai-completions or anthropic-messages compat API) that doesn't always fill in required tool parameters perfectly. These models were working fine before 2026.3.22.

Temporary Workaround

Use tools.byProvider to deny problematic tools for affected providers:

{
  tools: {
    byProvider: {
      "kimi-coding": { deny: ["exec", "process"] },
      "kimi": { deny: ["exec", "process"] }
    }
  }
}

This prevents the validation from being reached, but removes exec functionality for these models entirely.

Suggested Fix

In @mariozechner/pi-ai (or in the OpenClaw wrapper that calls validateToolArguments): catch validation errors and return a structured tool_error to the model instead of throwing, so the agent turn can continue.

Alternatively, expose a per-provider config option to skip or soften tool argument validation:

{
  models: {
    providers: {
      "kimi": {
        toolCallValidation: "warn"  // "strict" | "warn" | "off"
      }
    }
  }
}

References

Related to the introduction of @mariozechner/pi-ai in 2026.3.22
@mariozechner/pi-ai validation code: /node_modules/@mariozechner/pi-ai/dist/utils/validation.js

extent analysis

Fix Plan

To address the issue, we need to modify the validateToolArguments function in @mariozechner/pi-ai to catch validation errors and return a structured tool_error instead of throwing an exception. Here are the steps:

Modify the validateToolArguments function to catch validation errors:

export function validateToolArguments(tool, toolCall) {
    const validate = ajv.compile(tool.parameters);
    const args = structuredClone(toolCall.arguments);
    try {
        if (validate(args)) return args;
    } catch (error) {
        return { type: "tool_error", error: `Validation failed for tool "${toolCall.name}": ${error.message}` };
    }
}

Alternatively, expose a per-provider config option to skip or soften tool argument validation. This can be done by adding a new option to the models.providers config:

{
  models: {
    providers: {
      "kimi": {
        toolCallValidation: "warn"  // "strict" | "warn" | "off"
      }
    }
  }
}

Then, modify the validateToolArguments function to respect this config option:

export function validateToolArguments(tool, toolCall, providerConfig) {
    const validate = ajv.compile(tool.parameters);
    const args = structuredClone(toolCall.arguments);
    if (providerConfig.toolCallValidation === "off") return args;
    try {
        if (validate(args)) return args;
    } catch (error) {
        if (providerConfig.toolCallValidation === "warn") {
            // log a warning instead of returning an error
            console.warn(`Validation failed for tool "${toolCall.name}": ${error.message}`);
            return args;
        } else {
            return { type: "tool_error", error: `Validation failed for tool "${toolCall.name}": ${error.message}` };
        }
    }
}

Verification

To verify that the fix worked, test the validateToolArguments function with different input scenarios, including:

Valid tool arguments
Invalid tool arguments (e.g. missing required fields)
Tool arguments with unknown fields
Tool arguments with invalid field types

Verify that the function returns the expected output for each scenario, including the structured tool_error object when validation fails.

Extra Tips

Make sure to update the @mariozechner/pi-ai version to the latest version that includes the modified validateToolArguments function.
Consider adding additional logging or monitoring to track validation errors and improve the overall robustness of the system.
If using the per-provider config option, make sure to document the available options and their effects on the system behavior.

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 #request error #file not found #serialization error #model compatibility

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] @mariozechner/pi-ai tool validation throws on empty args instead of returning tool_error — regression from 2026.3.22 [2 comments, 1 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

Root Cause

Fix Action

Temporary Workaround

Code Example

Problem Description

Root Cause

Environment

Observed Behavior

Expected Behavior

Why This Matters

Temporary Workaround

Suggested Fix

References

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

TRENDING

openclaw - 💡(How to fix) Fix [Bug] @mariozechner/pi-ai tool validation throws on empty args instead of returning tool_error — regression from 2026.3.22 [2 comments, 1 participants]

Recommended Tools

GitHub issue graph ai analysis

Error Message

Root Cause

Fix Action

Temporary Workaround

Code Example

Problem Description

Root Cause

Environment

Observed Behavior

Expected Behavior

Why This Matters

Temporary Workaround

Suggested Fix

References

extent analysis

Fix Plan

Verification

Extra Tips

Still need to ship something?

RELATED_DISCOVERY

TRENDING