Feature: Tool Call Validation Hooks

Summary

Add configurable validation hooks that intercept tool calls before execution, enabling custom validation logic (timestamp verification, data format checks, permission enforcement).

Problem

Real-world Impact: Model Time Drift

In production use (2026-03-08 to 2026-03-09), the AI model calculated timestamps with wrong year (2025 instead of 2026), leading to 18 corrupted records in our Feishu Bitable:

Root Cause: AI models trained on historical data may produce outdated timestamps, especially for UTC millisecond calculations + timezone conversions.

Current Limitation

OpenClaw skills provide guidelines (SKILL.md) but cannot enforce validation before API calls. This is fragile: models may skip validation, especially in multi-turn conversations or when context is long.

Proposed API

Add toolValidationHooks to Gateway config:

{
  "toolValidationHooks": {
    "feishu_bitable_app_table_record": {
      "before": "/path/to/validate-timestamp.js",
      "onError": "reject" // or "warn" or "log"
    }
  }
}

Hook Script Interface

// validate-timestamp.js
module.exports = async function(toolName, params, context) {
  if (params.fields && params.fields['日期']) {
    const timestamp = params.fields['日期'];
    const year = new Date(timestamp).getFullYear();
    const currentYear = new Date().getFullYear();

    if (Math.abs(year - currentYear) > 1) {
      return {
        valid: false,
        error: `Invalid year: ${year}, expected ${currentYear}±1`,
        suggestion: `Use helper: node trade-time-helper.js "YYYY-MM-DD HH:mm:ss"`
      };
    }
  }

  return { valid: true };
};

Execution Flow

1. Agent calls tool (e.g., feishu_bitable_app_table_record)
2. Gateway intercepts → runs validation hook
3. If valid → proceed with tool execution
4. If invalid → reject with error message (or warn and continue)

Expected Impact

Benefits:

• Prevent corrupted data before it enters external systems
• Generic mechanism (not limited to timestamps)
• Per-tool or pattern-based configuration
• Non-breaking (opt-in via config)

Implementation Notes

Phase 1:

• Add toolValidationHooks config schema
• Implement hook execution in Gateway
• Support before hooks (post hooks can be Phase 2)
• Error handling: reject/warn/log modes

Phase 2 (optional):

• Async validation with timeout
• Hook chaining (multiple validators per tool)
• Context injection (session info, agent ID)

Related

• Related to Feature Request: Middleware hooks for agent protocol enforcement #30504 - Middleware hooks for agent protocol enforcement
• Related to Feature: Tool Result Middleware / Output Compression Hook #35053 - Tool result middleware / output compression hook
• Related to fix(cron): reject past timestamps in createJob (defense-in-depth) #18856 - Reject past timestamps in cron jobs

Environment

• OpenClaw: 2026.3.7
• Feishu Plugin: 2026.3.7-beta.1
• Models affected: GLM-4.7, GPT-4, Claude 3.5 (likely all LLMs)

Would the OpenClaw team be open to this feature? Happy to contribute implementation if the design direction is approved.