The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/predict_distribution.py utilizes pickle.load() and joblib.load() to deserialize model files provided via command-line arguments. Both functions are inherently unsafe as they allow for arbitrary code execution if the input file has been tampered with or is malicious.
[COMMAND_EXECUTION]: Several R scripts, including scripts/predict_distribution.R, scripts/project_scenarios.R, and scripts/run_ensemble_sdm.R, use readRDS() to load model objects. In R, readRDS can execute embedded expressions during the loading process, posing a risk of arbitrary code execution from malicious data files.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface by ingesting untrusted external data.
Ingestion points: scripts/run_ensemble_sdm.R and scripts/sdm_pipeline.py read occurrence data from CSV files and parameters from YAML files.
Boundary markers: The instructions lack explicit boundary markers or "ignore instructions" warnings for the data being processed.
Capability inventory: The skill scripts have the capability to write various files (GeoTIFF, CSV, PNG) and generate a methodological narrative (sdm_report.md) that incorporates user-provided metadata like species names.
Sanitization: There is no evidence of sanitization or escaping of the input data before it is interpolated into reports or used to drive modeling decisions.

species-distribution-modeling