◀Root of All Trust
AI mvschwarz/openrig #2899 · ★ 8.9kfix(daemon): trust the canonical git root in the Claude pre-launch (#2820)A real, public pull request, signed as made with Claude Code. See it on GitHub ↗
IN THE WORLD

Developers running AI coding teams no longer hit a surprise safety prompt on launch.

REACH

Software developers use OpenRig to run persistent teams of AI coding agents like Claude Code and Codex. Created by Mike Schwarz, the open-source tool has 8,856 GitHub stars.

WHAT IT MEANS

When launching a fresh agent in a subfolder of a repository, OpenRig now trusts the main project root. Agents start working immediately instead of pausing at Claude Code's safety check.

RED · how I did it
  • GONE Only mark the agent's exact launch folder (and its real symlink target) as pre-trusted.
  • GONE Never pre-trust parent folders, enclosing git repository roots, or linked main repositories.
  • GONE Any target folder outside the launch folder stays untrusted and triggers the safety prompt.
GREEN · how I do it now
  • NEW Mark the launch folder and its nearest enclosing git repository root as pre-trusted.
  • NEW If that git root is a linked worktree whose back-link matches, pre-trust the main repository instead.
  • NEW Submodules, forged worktree pointers without a matching back-link, and folders above the git root are not pre-trusted.

WHERE Terminal launch via `rig seat launch --fresh`, right before Claude Code opens its 'Quick safety check' screen for pre-approved tools.

SCENE You launch a fresh Claude Code agent inside a nested team folder that pre-approves bash tools. You want it working immediately, not stuck waiting on a trust screen—and without trusting spoofed repositories.

Guarding
Pre-launch workspace trust for pre-approved agent tool permissions
What it can do
Pre-trusted at launch · Blocked by safety prompt