Repository navigation
Conversation
Devin's "Do you trust the authors of this directory?" menu is a numbered select. Orca PTY input (stdin-after-start follow-up, draft paste) lands on that menu and answers "No, exit", so creating a Devin agent dies immediately. --permission-mode bypass only auto-approves tools. Pass --respect-workspace-trust=false on the launch command, the same pattern as command-code --trust, including manual-permission and resume launches. New Orca worktrees stay untrusted in Devin's store; the flag is per-invocation. Fixes stablyai#15859
|
No actionable comments were generated in the recent review. 🎉 ℹ️ Recent review info⚙️ Run configurationConfiguration used: Repository UI Review profile: CHILL Plan: Pro Plus Run ID: 📒 Files selected for processing (4)
Included review availability: Your plan provides up to 10 included reviews per hour; 9 remain after this review. 📝 WalkthroughWalkthroughThe Devin launch command now includes 🚥 Pre-merge checks | ✅ 5✅ Passed checks (5 passed)
Thanks for using CodeRabbit! It's free for OSS, and your support helps us grow. If you like it, consider giving us a shout-out. Comment |
|
thanks for reporting. looking into it |
|
Thanks @dbachko for the clear root-cause analysis in #15859! This is now fixed on main by #21925 (ac4d6b4), which adds Devin's documented |
ELI5
Creating a Devin agent in Orca was dying immediately. Devin asks "do you trust this folder?" with a numbered menu, and Orca's first keystrokes into the terminal were answering No, exit. This change tells Devin not to show that menu when Orca launches it.
What Changed
Devin's launch command is now
devin --respect-workspace-trust=false(still followed by--permission-mode bypassin YOLO mode). That is Devin's documented skip for the first-run workspace-trust prompt, the same pattern ascommand-code --trust.The flag lives on
launchCmd, not YOLO args, so manual permission mode and session resume also skip the menu. New Orca worktrees are not written into Devin's global trust store; the skip is per invocation.Why
--permission-mode bypassonly auto-approves tools. It does not skip workspace trust. Devin usesstdin-after-start, so as soon asdevinis the foreground process — which it already is while the trust menu is up — Orca's follow-up prompt or draft paste is consumed as a menu choice. Option 2 is "No, exit".Orca already pre-writes trust artifacts for Cursor, Copilot, and Codex (
preflightTrust). Devin has a public CLI flag for this, so the launch command is the smaller, more reliable fix than reverse-engineeringtrusted_workspaces.json.Linked Issue
Fixes #15859
Related: #11765 (Cursor trust auto-quit), #10666 (inject answers Codex trust prompt)
Visual Proof
N/A — launch-command change only. No UI, layout, or interaction surface.
Testing
Ran:
98 passed.
Verified on this machine that
devin --respect-workspace-trust=false -p "…"runs in an untrusted directory instead of failing the trust check (Devin CLI 3000.5.20). Did not re-create a Devin agent inside a running Orca app.Platforms: macOS unit tests. The flag is Devin CLI argv, so Linux, Windows, WSL, and SSH hosts pick it up as long as that host's Devin supports
--respect-workspace-trust(current docs / 3000.x). Older Devin without the flag would fail at clap parse; Devin auto-updates by default.Review
=falseis one token sofalsecannot be parsed as a PATH.--permission-mode bypass; the trust skip is independent so toggling manual/yolo cannot reintroduce the menu.skip_workspace_trustin~/.config/devin/config.json, so a Devin run outside Orca still prompts.Agent skill upstream boundary
docs/reference/agent-skill-sharing-upstream-boundary.mdand copies or mechanically translates no upstream skill-installer source, tests, fixtures, registry entries, path tables, comments, or documentation.Notes
Custom command overrides that replace the whole launch with bare
devinwill drop the flag, same ascommand-code --trusttoday.Checklist
N/Awith reasonpnpm lint,pnpm typecheck,pnpm test, andpnpm buildpass (or CI will cover; local preferred)