Stage 7 · Lesson 2 of 5
Permission levels and tool approval in Copilot
In chat, the permissions picker sets how much an agent session may do. Manual uses your configured approval settings. Assisted has a model judge each tool call and asks you when it does not approve.
all every tool call, and VS Code warns it should not be treated as normal. There is also an Autopilot mode that auto-approves and keeps iterating until it decides the task is done. Agents can save file edits without asking even in Manual, so review diffs.
In the , --allow-tool and --deny-tool set rules, such as --deny-tool='shell(git push)', and deny always wins; --allow-all-tools gives it the same access you have.
Common traprunning with auto-approval on a you have not read.
A question to check yourself, an AI practice session and 2 cheat sheet entries come when you add the roadmap to your account. Add it
Resources for this stage
Start with
- GuideFreeBuilding Effective Agents
Anthropic · Advanced
Anthropic's clear, free essay on how agents work and when simpler workflows are better
Keep handy
Also worth your time
Permission levels, auto-approval risks, sandboxing and the security baseline for agents
- OWASP Top 10 for LLM ApplicationsGo deeper
Prompt injection and excessive agency explained by the security community