openclaw - 💡(How to fix) Fix [Proposal] Self-Healing Execution: 解决工具调用死循环问题 [1 participants]

1. 工具层：输入约束严格，模型很容易产生 validation error
2. Validation error、schema mismatch、missing field 这类确定性错误，禁止无限重试 | 可自动修复错误 | validation error、provider not found、gateway 未启动 | 进入 self-heal 流程 |
3. Error Classifier — 判断 retryable / self-healable / non-retryable 收到请求 → 执行动作 → 报错 → Error Classifier →
4. 工具参数自修 — validation error 后重新读取 schema，只保留 required 字段

• Validation error 一律禁止直接重试

1. Classify the error first.
2. If the error is deterministic (validation error, schema mismatch, missing required field, invalid enum, bad config reference, provider not found), do not retry the original action directly.
3. If the same class of error happens again in the same turn, stop and return a structured failure summary.

• 短期：止血，禁止 deterministic error 无限重试

TL;DR

Implement a self-healing mechanism that classifies errors, generates repair plans, and validates repairs before re-running original actions to prevent infinite retries.

Guidance

• Identify and classify errors into retryable, self-healable, and non-retryable categories to determine the best course of action.
• Implement a repair planner to generate repair actions for self-healable errors, and a verifier to validate the success of these repairs.
• Preserve execution state outside of compacted conversation history to prevent loss of failure information and enable more informed retry decisions.
• Limit the number of repair attempts and retries to prevent infinite loops and implement a circuit breaker to stop execution when repair attempts exceed a threshold.

Example

A simple example of error classification and repair planning could involve checking the error type and generating a repair plan accordingly, such as:

if error_type == "validation_error":
    repair_plan = generate_repair_plan_for_validation_error()
elif error_type == "provider_not_found":
    repair_plan = generate_repair_plan_for_provider_not_found()

Notes

The implementation of the self-healing mechanism will require significant changes to the existing system, including the addition of new modules and the modification of existing execution flows. It is essential to carefully consider the design and implementation of each component to ensure that the system can effectively classify errors, generate repair plans, and validate repairs.

Recommendation

Apply the workaround of implementing a self-healing mechanism with error classification, repair planning, and validation to prevent infinite retries and improve system reliability. This approach will allow the system to recover from errors more effectively and reduce the likelihood of repeated failures.