Sync from GitHub via hub-sync

AGENTS.md

@@ -77,6 +77,19 @@ A couple of scripts have `-bucket.py` variants (currently `falcon-ocr-bucket.py`
 `ls` the repo to check whether a `-bucket.py` variant exists for the model you want
 before assuming it's available.
 
+The `-v` source can also be a **local directory** (`huggingface_hub` ≥ 1.22) — the CLI
+syncs it to a private bucket and mounts it, so local images/PDFs need no upload step:
+
+    hf jobs uv run --flavor l4x1 -s HF_TOKEN \
+      -v ./pages:/input \
+      -v hf://buckets/<user>/<output>:/output \
+      https://huggingface.co/datasets/uv-scripts/ocr/raw/main/<script>-bucket.py \
+      /input /output
+
+Re-runs only sync new/changed files. Local mounts are read-only by default; a local
+*output* dir works too with `:rw` — pull results back with the `hf buckets sync`
+command the CLI prints at launch.
+
 ## Common flags across dataset-mode scripts
 
 Most scripts support: `--max-samples`, `--shuffle`, `--seed`, `--split`, `--image-column`,

CLAUDE.md

@@ -26,7 +26,8 @@ planned **self-review skill** (see [Deferred](#deferred--tracked)) just enforces
   dep (e.g. bucket I/O → `bucketbag`, [#67](https://github.com/davanstrien/uv-scripts-for-ai/issues/67)),
   never a local import. Recipes = inline; internal tooling may share freely.
 - **GPU + output.** Check `torch.cuda.is_available()` and exit clearly if absent. Write to the Hub
-  (`push_to_hub`) or a bucket (`-v hf://…`), never local paths (Jobs disk is ephemeral).
+  (`push_to_hub`) or a bucket (`-v hf://…`), never bare local paths (Jobs disk is ephemeral; local *input*
+  dirs mount via `-v ./dir:/mnt`, synced to a bucket automatically).
 - **vLLM image + fail-fast preflight.** If the model's arch isn't in a stable vLLM wheel, **omit
   `vllm`/`torch` from deps** and run on the pinned `vllm/vllm-openai` image via
   `--image … --python … -e PYTHONPATH=…`; add a preflight that `sys.exit(1)`s naming the exact flags