Five ways an agent can fake a green build
Agents rarely cheat on purpose, but under pressure to make tests pass they find shortcuts. Here is what to look for — and what PaneScale flags automatically.
Skipping an agent’s permission prompts saves hours — and occasionally costs a lot more. When it is reasonable, and what to put in place first.
Every coding CLI asks before it does something risky: run this command? edit that file? install this package? Every CLI also has a switch to stop asking. In Claude Code it is --dangerously-skip-permissions, in Codex --dangerously-bypass-approvals-and-sandbox, in Gemini CLI --yolo. The names are honest.
PaneScale lets you start panes with these flags (Settings → Agents & accounts → Start without prompts). This post is about when you should.
Permission prompts are a human in the loop at the moment it matters most — right before an action. Without them, the agent can:
Most of the time it does none of that. The question is what happens the one time it does.
| Situation | Bypass? | Why |
|---|---|---|
| Isolated lane, tests and type checker as finish line, no secrets in the repo | Reasonable | Damage stays in a throwaway worktree, and you review the diff anyway |
| Overnight autopilot with limits and required checks | Reasonable | Nobody would answer the prompts; the guardrails do the work instead |
| Your main working copy with uncommitted work | No | One wrong command and the work is gone |
| Repository with deploy scripts or production credentials | No | A “helpful” deploy is one command away |
| Tasks that need the network beyond package installs | Think twice | Exfiltration risks grow with every reachable host |
| You’re watching the pane anyway | Unnecessary | Prompts cost you a keypress |
If you decide to bypass, don’t bypass alone:
Panes started without prompts carry a marker in their header. That sounds minor, but it matters when you have nine panes open and want to know at a glance which ones you can let run.
Bypass is a good default for isolated, verifiable, unattended work. It is a bad default for anything that touches your main working copy, your credentials or your production systems. When in doubt, keep the prompts and answer them from the Attention dock.
Agents rarely cheat on purpose, but under pressure to make tests pass they find shortcuts. Here is what to look for — and what PaneScale flags automatically.
The difference between a task that runs overnight and one that wanders is usually a single sentence — the finish line. A short template and examples.
When agents work in multi-step workflows, “it worked” isn’t enough. What a run record should contain, and how PaneScale’s workflow runs provide it.