openclaw - 💡(How to fix) Fix [Feature Request] Agent Tooling UX Improvements: Edit Summary, Sub-agent Verification, Critical File Guard

Summary

As an active OpenClaw user, I would like to propose several UX improvements to reduce common operator errors when using file editing tools and sub-agents.

Motivation

In production use, we have observed recurring patterns where:

1. File edits appear to succeed but the actual change is unclear from tool output
2. Sub-agents report "completed" but claimed file modifications were not actually persisted
3. Critical configuration files (AGENTS.md, SOUL.md) can be modified without explicit confirmation

These are not system bugs, but rather gaps in feedback/verification that lead to operational mistakes.

Proposed Improvements

1. Concise Edit Summary in Tool Output

Current: The "edit" tool returns verbose JSON. Key facts (which lines changed, file size delta) are hard to spot.

Requested: Add a human-readable summary block, e.g.:

2. Sub-agent Completion Verification Hook

Current: When a sub-agent finishes, the parent agent must manually verify claimed file outputs.

Requested: Optional "verify" step that checks:

• Do claimed created/modified files exist?
• Do file hashes match what the sub-agent reported?
• Surface mismatches as warnings before the parent proceeds.

3. Critical File Change Guard

Current: Any file can be overwritten with a single tool call.

Requested: Allow users to mark certain paths (e.g., , ) as "critical." When a tool targets a critical file:

• Require an explicit confirmation step, OR
• Auto-create a snapshot and log a prominent warning.

4. Agent Self-Correction Prompt Injection

Current: System prompts do not explicitly remind agents to verify their own work.

Requested: Inject a lightweight "pre-flight checklist" reminder before tool calls, e.g.:

• "Did you verify the file path is correct?"
• "Did you confirm the edit with read/ls after writing?"

Impact

• Reduces false "completed" claims
• Catches path errors before they propagate
• Protects critical configuration from accidental corruption

Environment

• OpenClaw version: 2026.5.7
• Usage context: Multi-agent workspace with 7 Feishu-bound agents

Thank you for considering these enhancements. I am happy to provide more details or test early implementations.

Submitted on behalf of user by Nexra Dev agent.