feat: add --dangerously-skip-permissions flag

[surma]

I don't know if you all have any interest in this, but I whipped this up because I really enjoy the simplicity of this flag in Claude Code.

So this adds a CLI flag and environment variable to bypass all permission prompts, similar to Claude Code's implementation.

Happy to take feedback or just throw this away if you don't think this is the right approach.

Usage

# CLI flag
opencode --dangerously-skip-permissions
# Environment variable
OPENCODE_DANGEROUSLY_SKIP_PERMISSIONS=true opencode
# With run command
opencode run --dangerously-skip-permissions "do something"

Behavior

• Skips all ask permission prompts (auto-allows)
• deny rules still apply (explicit denies are respected)
• Shows "△ YOLO mode" warning in the TUI (home screen, footer, and sidebar)

Changes

• flag.ts: Add OPENCODE_DANGEROUSLY_SKIP_PERMISSIONS env var
• permission/next.ts: Add skipPermissions state and bypass logic in ask()
• run.ts: Add --dangerously-skip-permissions option
• thread.ts: Add --dangerously-skip-permissions option, propagate to TUI via args
• args.tsx: Add dangerouslySkipPermissions to Args interface
• home.tsx, footer.tsx, sidebar.tsx: Show warning indicator when enabled

Fixes#8463

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

No duplicate PRs found

[github-actions]

Hey! Your PR title Implement --dangerously-skip-permissions aka YOLO mode doesn't follow conventional commit format.

Please update it to start with one of:

• feat: or feat(scope): new feature
• fix: or fix(scope): bug fix
• docs: or docs(scope): documentation changes
• chore: or chore(scope): maintenance tasks
• refactor: or refactor(scope): code refactoring
• test: or test(scope): adding or updating tests

Where scope is the package name (e.g., app, desktop, opencode).

See CONTRIBUTING.md for details.

[mguttmann]

For reference — I've implemented a more comprehensive YOLO mode in PR #11833 that builds on this concept with additional safety and UX:

• ✅ Respects explicit deny rules (only auto-approves ask rules)
• ✅ Session-only mode (resets on restart) + persistent mode (saved to config.json)
• ✅ Full Settings UI with danger warnings
• ✅ Env var support (OPENCODE_YOLO=true)
• ✅ CLI flag support
• ✅ All CI checks passing

Tracking issue: #11831

[phillmorgan28]

any progress on this?

[Winter-Dry]

I really need this feature, when can ths feature get merged!

[dingkwang]

Looks good!

[fiesh]

Any news on this? It's quite annoying that there is no way to disable rather pointless security mechanism. Running the server in a podman container should be the default anyway, and there it's basically superfluous.

[fiesh]

Any news on this? It's quite annoying that there is no way to disable rather pointless security mechanism. Running the server in a podman container should be the default anyway, and there it's basically superfluous.

Actually, it seems I have found a way around this: specify --env HOME=/ when creating the container. This seems to circumvent all "external directory" considerations. I would have loved to have this documented somewhere!

[fiesh]

Any news on this? It's quite annoying that there is no way to disable rather pointless security mechanism. Running the server in a podman container should be the default anyway, and there it's basically superfluous.

Actually, it seems I have found a way around this: specify --env HOME=/ when creating the container. This seems to circumvent all "external directory" considerations. I would have loved to have this documented somewhere!

Actually, that also doesn't seem to work... sigh.

[fiesh]

Any news on this? It's quite annoying that there is no way to disable rather pointless security mechanism. Running the server in a podman container should be the default anyway, and there it's basically superfluous.

Actually, it seems I have found a way around this: specify --env HOME=/ when creating the container. This seems to circumvent all "external directory" considerations. I would have loved to have this documented somewhere!

Actually, that also doesn't seem to work... sigh.

Ok so it works out of the box with a fresh config, without any HOME being set, and then same agents screw it up. So I guess it was all correct all along, and one has all permissions in a container as root. 👍

[mystery4f]

When will this be available for use? I really want it.

[surma]

This PR is prob wildly outdated now. I am sure one of the many other ones can land :)