The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the installation of the jinko-sdk library with the nonmem2jinko extra. As this package is a primary resource from the vendor (Nova In Silico) used to fulfill the skill's documented purpose, it is considered a legitimate dependency.
[COMMAND_EXECUTION]: To verify model accuracy, the compare_against_reference.py script executes local commands via Rscript to leverage specialized R packages such as nonmem2rx and rxode2. These operations are limited to numerical validation and represent standard scientific workflow capabilities.
[INDIRECT_PROMPT_INJECTION]: The conversion process involves ingesting untrusted external data in the form of NONMEM control streams (.mod, .ctl, .lst) and CSV data sets.
Ingestion points: User-provided files are loaded in convert_model.py and convert_trial.py to be processed as pharmacokinetic models.
Boundary markers: The skill explicitly instructs the agent to use automated scripts and strictly prohibits manual translation of code blocks, which serves as a procedural guardrail to prevent the agent from executing instructions embedded within the model files.
Capability inventory: The skill possesses capabilities to read local files, interact with the Jinkō API, and execute local validation tools (R).
Sanitization: Input data is parsed and transformed by a transpiler library (nonmem2jinko) that converts imperative code into declarative parameters, preventing direct interpolation of raw input into the agent's core instruction set.

jinko-task-from-nonmem