The Agent Skills Directory

[SAFE]: The skill performs its stated medical research function without any detected malicious patterns. It lacks network communication, credential harvesting, or unauthorized privilege escalation mechanisms.
[COMMAND_EXECUTION]: The skill executes modular R scripts (e.g., scripts/main.R, scripts/wgcna_core.R) using standard relative path resolution via get_script_dir(). This is an expected and safe implementation for a multi-script local analysis pipeline.
[INDIRECT_PROMPT_INJECTION]: The skill possesses a data ingestion surface as it reads user-provided CSV files (input_file, group_file). However, it mitigates risks through strict validation:
Ingestion points: Files are read using data.table::fread in scripts/io.R and scripts/chunk_io.R.
Boundary markers: The agent is instructed to validate inputs in SKILL.md.
Capability inventory: Capabilities are limited to local file read/write and R computation.
Sanitization: scripts/chunk_io.R implements coerce_expression_chunk which enforces numeric values for expression data, and load_group_data in scripts/io.R validates sample/group column integrity, preventing non-data payloads from reaching sensitive execution contexts.
[DYNAMIC_EXECUTION]: The skill uses saveRDS and loadRDS in scripts/io.R for internal data persistence between analysis steps. While RDS can be a vector for object injection, here it is used strictly within the skill's own defined output_dir for its own generated results, representing low risk in this closed execution loop.

wgcna-analysis