Troubleshooting

Why Your Agent Says “Approval Required”

What that message means, when it’s expected, and exactly what to do so you can keep moving without confusion.

If you're new to OpenClaw/heyron, this is one of the most confusing moments: you ask your agent to do something, and it replies with an approval request instead of just doing it.

Short version: nothing is broken. Approval is a safety feature. Your agent is asking for permission before running a sensitive command (like changing system settings, using elevated permissions, or touching risky files).

What “approval required” actually means

Agents can do a lot, but they should not silently run high-risk commands. When the system detects a potentially sensitive action, it pauses and asks you to approve the exact command first.

Common triggers

Installing system packages or tools
Commands that require admin/sudo-level privileges
Potentially destructive operations (delete/move/overwrite)
Commands that access protected locations outside the normal workspace

How to approve safely

1) Read the full command first

Don't approve blindly. Confirm it matches what you asked for and doesn't include extra unexpected operations.

2) Use the exact approval command provided

Your agent will usually give you a one-liner (for example with /approve allow-once ...). Copy and run exactly that if it looks correct.

3) Preserve the full command (including &&, pipes, and flags)

A common mistake is approving a shortened or rewritten version. If the prompt includes command chains, pipes, semicolons, or multiline script text, approve the full command exactly as shown so the system can match it safely.

4) Prefer allow-once unless you're sure

One-time approval is safer than broad approval. It's usually enough to finish the task while keeping guardrails strong.

Important: if a second sensitive command appears later, you may need to approve again. That's expected behavior.

When you should deny

The command is unrelated to your request
You see destructive flags you didn't ask for
The command chains many actions you don't understand
It tries to access data or directories that feel out-of-scope

When in doubt, deny and ask the agent to explain each part of the command in plain language.

Fast troubleshooting checklist

Confirm the command shown matches your request
Use the exact /approve ... line from the agent output (don't paraphrase)
If the command includes &&, |, ;, or multiline text, approve the complete version exactly as shown
Approve once (not always) for the immediate action
If it asks again, review the new command — don't assume it's identical
If it keeps looping on approvals, ask the agent to switch to a safer non-elevated approach or split work into smaller commands

Common real-world gotchas

Copied the wrong line: In busy chats, users sometimes copy explanatory text instead of the approval command. Only run the /approve ... command line.
Edited the command before approving: Even small edits can cause mismatch and another approval request. Re-copy the original approval line and retry.
Second approval appears later: This is normal when the first approved step leads to another sensitive step. Review and approve each one intentionally.

Example: good vs risky

✅ Good approval candidate

Install ripgrep so we can search files faster: brew install ripgrep

🚫 Risky without explanation

sudo rm -rf /some/path && curl ... | sh

Best practice: ask your agent to break complex commands into smaller steps. Smaller steps are easier to review and safer to approve.

Bottom line

Approval prompts are not friction for the sake of friction — they protect you from accidental damage. Once you get used to reviewing and approving intentionally, you'll move faster and safer.