Troubleshooting

Approval Keeps Coming Back?

When it feels like you approved already but the agent still asks again, it’s usually a command mismatch or a new step. Fix it fast.

This is one of the most frustrating loops in support: users approve once, then get another approval prompt and assume the system is broken. Most of the time, it’s doing the right safety behavior.

Bottom line: allow-once approves one exact command only. If the next step changes, it needs a new approval.

Why approval prompts repeat

New command: the agent moved to a different command than the one you approved.
Command mismatch: edited/copy-pasted approval line changed operators, quotes, or flags.
Multi-step workflow: step 1 approved, step 2 still pending.
Different elevation: safe command ran, but elevated command still needs approval.

2-minute recovery workflow

Read the full pending command carefully.
Use the exact approval line the agent/tool provided.
Do not paraphrase or trim chained parts (&&, pipes, semicolons, multiline).
If another approval appears, treat it as a new command and repeat safely.

Known-good approval behavior

✅ Approve exactly what is shown: /approve allow-once [full command exactly as provided] ❌ Avoid this: /approve allow-once [shortened/summarized version]

Pro tip: if the task clearly needs several commands, choose allow-always only when you trust the scope and source. Otherwise keep using allow-once per step.

High-friction mistakes to avoid

Approving one command and assuming all future commands are covered.
Removing operators like && from the approval line.
Approving a “safe” command but ignoring the next elevated one.
Switching sessions/threads and expecting previous approvals to carry over.
Approving a command from an older message after the tool output changed.

When `allow-always` still asks again

This confuses people a lot: allow-always can still trigger a new approval prompt if the next command is materially different or has a different risk profile.

Different command body: added flags, changed targets, or a new chained command block.
Different privilege level: a normal command ran, then a separate elevated command appears.
Different command context: new shell/script wrapper or multi-line script generated after retries.

If prompted again, don’t assume failure. Compare old vs new command text and approve only what you intend to run.

Important: always approve using the exact /approve ... line shown in the current prompt. Do not reuse a previous approval line from memory.

Copy/paste support packet for #help

Approval loop issue Task: Command I approved (exact): New command that requested approval: Did I use allow-once or allow-always?: Any error text shown: Where this happened (dashboard/Discord/thread):