diff --git a/.claude/skills/ptq/references/slurm-setup-ptq.md b/.claude/skills/ptq/references/slurm-setup-ptq.md
index 234b03c5e0..038823ba2e 100644
--- a/.claude/skills/ptq/references/slurm-setup-ptq.md
+++ b/.claude/skills/ptq/references/slurm-setup-ptq.md
@@ -28,10 +28,16 @@ If enroot import fails (e.g., permission errors on lustre), use pyxis inline pul
 
 ### Container dependency pitfalls
 
-**New models may need newer transformers** than what's in the container. Install from source inside the job script:
+**New models may need newer transformers** than what's in the container. Install from PyPI inside the job script (unset `PIP_CONSTRAINT` first if needed — see below):
 
 ```bash
-pip install git+https://github.com/huggingface/transformers.git --quiet
+pip install -U transformers
+```
+
+Only install from git if the fix you need isn't in a released version yet:
+
+```bash
+pip install git+https://github.com/huggingface/transformers.git
 ```
 
 **Prefer `PYTHONPATH`** to use the synced ModelOpt source instead of installing inside the container — this avoids risking dependency conflicts (e.g., `pip install -U nvidia-modelopt[hf]` can upgrade PyTorch and break other packages):
@@ -90,4 +96,4 @@ This catches script errors cheaply before using GPU quota on a real run.
 
 See `skills/common/slurm-setup.md` section 2 for the smoke test partition pattern.
 
-Only submit the full calibration job after the smoke test exits cleanly.
+Only submit the full calibration job after the smoke test exits cleanly.
\ No newline at end of file