[webtecnica]

Summary

On macOS, TCC permission grants (microphone, screen recording, etc.) are keyed to the resolved client binary path. The uv-managed Python interpreter lives in ~/.local/share/uv/python/cpython-<patch>-macos-*/bin/python3.11 — every patch bump creates a new directory → orphaned grants → permission prompts return after every update. Symlinks don't help (TCC resolves through them).

Change (stable anchor — "copy to fixed path before invoking")

• hermes_cli/macos_tcc_anchor.py (new, +277): replaces the venv/bin/python symlink (which resolves into the versioned uv store) with a real copy of the interpreter binary. The venv path is stable across updates and, being a regular file, there's no symlink for TCC to resolve → constant client path. pyvenv.cfg (home → uv store) untouched, so stdlib still loads as today. Self-healing: detects patch bump via .tcc-anchor-source marker (re-copies) and re-points python3/python3.11 aliases to the anchor. No-op on non-macOS and non-uv interpreters (Homebrew already has a stable path). Best-effort, never raises.
• hermes_cli/doctor.py (+52): reports anchor state (healthy / stale / would-fix).
• hermes_cli/update_cmd.py (+15): runs the anchor heal after update.
• tests/hermes_cli/test_macos_tcc_anchor.py (new, +296): uv-store detection, patch-bump re-copy, symlink replacement, no-op cases (monkeypatched — real TCC behavior not testable on Linux CI).

Verification

• tests/hermes_cli/test_macos_tcc_anchor.py: 20 passed. Real macOS validation left to review (CI is Linux).

Closes #85345

[Enough1122]

AI code review — automated review for reference, author can ignore or act on any point.

fix(macos): stable TCC anchor for uv-managed python interpreter

Good design (real-file copy + pyvenv.cfg home keeps the stdlib source; atomic temp+rename; best-effort with doctor/update integration). Points to consider:

• Test gap: tests/hermes_cli/test_macos_tcc_anchor.py exercises fake text "binaries" only. Real-binary concerns — code-signature preservation after shutil.copy2 (macOS requires at least ad-hoc signatures) and dylib/framework resolution from the new venv path — are untested. Recommend a smoke test on real macOS hardware (venv/bin/python -c "import sys" after anchoring, and again after a simulated patch bump).
• _SIBLING_NAMES / _STORE_BIN_NAMES hardcode 3.11–3.13. A uv-managed 3.10 or 3.14+ interpreter leaves its python3.x alias symlink resolving into the versioned store — anything spawned through the alias still churns TCC. Consider deriving the version from the resolved interpreter instead of a fixed list.
• ensure_tcc_anchor reads _interpreter_source and then os.replaces the copy; if a concurrent uv sync rewrites bin/python in between, the anchor can pin a stale copy. The marker-based self-heal on the next hermes update / doctor --fix covers it, but re-reading the source immediately before replace (or comparing the marker first) would close the window.

[teknium1]

Merged via #95131 (5259f56) with your commit cherry-picked intact — thanks @webtecnica! Follow-up on top: alias/store binary names are now derived from the running interpreter (with a glob fallback) instead of the hardcoded 3.11–3.13 lists, so future Python bumps can't leave an alias resolving back into the versioned store. The real-Mac smoke test (grant surviving a simulated patch bump) is flagged for maintainer verification on hardware.