Launch Claude Code in auto mode - #5
lluisinthedesert wants to merge 2 commits into
Conversation
![devin-ai-integration[bot]](https://avatars.githubusercontent.com/in/811515?s=60&v=4){kind=small}
| run claude --permission-mode auto --plugin-dir "$SOURCE_DIR" "$PROMPT" | ||
| else | ||
| exec claude --plugin-dir "$SOURCE_DIR" "$PROMPT" | ||
| exec claude --permission-mode auto --plugin-dir "$SOURCE_DIR" "$PROMPT" |
There was a problem hiding this comment.
🟨 AI agent launched with automatic permission approval bypasses user confirmation for all actions
The installer now launches Claude Code with --permission-mode auto (install.sh:261), which automatically approves all permission requests (file writes, shell commands, network access, etc.) without user confirmation. The only constraint on Claude's actions is the natural-language prompt instruction to "not edit application code or call a model provider" (install.sh:257), which is not an enforceable security boundary. If the agent misinterprets the prompt or encounters an adversarial prompt injection in scanned files, it could perform destructive file operations or execute arbitrary commands with the user's full privileges.
Was this helpful? React with 👍 or 👎 to provide feedback.
1 participant
What changed
--permission-mode autoLMABbeneath the full ASCII wordmark in the installer and direct CLI--no-launchfor manual controlWhy
The one-line installer should hand off directly into the audit without a noisy sequence of routine permission prompts. Claude Code auto mode keeps its classifier in control while making the first-run flow substantially cleaner. The terminal lockup now teaches the LMAB acronym at the first visual beat.
Verification
--permission-mode autonpm run check— 6/6 tests passgit diff --check