Update data: 5 units

- legal/legal_dr_fees_01/base/software
- life_sciences/WGS_Variant_Calling/base/input
- life_sciences/amber_minimization_script_prep_instance_1/base/input
- life_sciences/amber_minimization_script_prep_instance_1/base/software
- life_sciences/amber_three_stage_mmgbsa_workflow_instance_1/base/input

tasks/legal/legal_dr_fees_01/base/software/README.txt

@@ -0,0 +1 @@
+This task uses Python 3.12 and open-source PDF libraries. Stage 4 verifies software/python3.12.

tasks/legal/legal_dr_fees_01/base/software/python3.12

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6055beb2f9c67330cea3f53c4e2252eeaa8d01d6e2689d47d35a58896aaec52e
+size 68

tasks/life_sciences/amber_minimization_script_prep_instance_1/base/software/README.md

@@ -0,0 +1,23 @@
+# Software Notes — amber_minimization_script_prep_instance_1
+
+This benchmark is a parser-only Amber minimization workflow-construction task.
+The evaluated agent produces three text files under the task output directory:
+
+- `leap.in`
+- `step2_implicit.mini.mdin`
+- `submit_min.sh`
+
+It does not execute `tleap`, `pmemd.cuda`, `MMPBSA.py`, `sbatch`, or
+`nvidia-smi`. Amber 22, CUDA 11.6.2, and SLURM appear in the authored
+`submit_min.sh` as the intended target environment only; they are not
+benchmark-provided at solve time on this cpu-free-ubuntu VM.
+
+## Agent-facing runtime contract
+
+- Linux shell (bash)
+- Python 3.10 at `/usr/bin/python3`, also reachable via this folder as
+  `software/python`
+- `uv` at `/usr/local/bin/uv` for evaluator-side tooling
+
+No other runtime is provisioned. Anything beyond the three text deliverables
+is out of scope for this task.