The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from a user-provided CSV file that is then processed by scripts/sensitivity_analysis.R and scripts/sensitivity_analysis.py.
Ingestion Point: The 'study' identifiers and numeric data from the input CSV are read into memory.
Boundary Markers: Absent. The instructions do not specify any delimiters or safety markers to differentiate between the data being processed and instructions to the agent.
Capability Inventory: The skill is capable of performing file system operations (reading CSV, writing PNG/CSV files) and outputting textual summaries to the console.
Sanitization: Absent. While numeric values are validated and parsed, the textual 'study' column is printed directly to the console as part of the final summary. This provides an opportunity for an attacker to include malicious instructions in the study names that could influence the agent when it reads the script output.
[COMMAND_EXECUTION]: The skill instructions direct the agent to execute shell commands (Rscript and python) using user-supplied paths for the input CSV and output directory.
Evidence: The SKILL.md documentation specifies a workflow calling Rscript scripts/sensitivity_analysis.R "<csv_path>".
Context: While the instructions use double quotes around variables, there is a potential risk of command injection if the agent or execution platform fails to properly escape shell metacharacters provided in the user-controlled path strings.

meta-sensitivity-plot