# TensorFlow SavedModel TextLineDataset PoC

This directory contains a restrained proof-of-concept TensorFlow SavedModel for a Huntr MFV submission.

The model reads a controlled local marker file at inference time:

```text
/tmp/tf_savedmodel_textlinedataset_secret.txt
```

It does not perform network access, destructive file writes, or shell command execution.

## Reproduce

```bash
python scripts/generate_tf_savedmodel_textlinedataset_poc.py
printf 'huntr-controlled-secret-line-1\nhuntr-controlled-secret-line-2\n' > /tmp/tf_savedmodel_textlinedataset_secret.txt
python - <<'PY'
import tensorflow as tf

model = tf.saved_model.load("artifacts/tf_savedmodel_textlinedataset_bypass/dataset_read_savedmodel")
print(model.signatures["serving_default"]()["contents"].numpy().decode())
PY
modelscan -p artifacts/tf_savedmodel_textlinedataset_bypass/dataset_read_savedmodel -r json
```