The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill ingests and processes external data files (sgRNA count matrices and sample sheets) provided by the user, which presents a surface for indirect prompt injection if the data contains malicious instructions aimed at influencing the agent's behavior during analysis.
Ingestion points: Data enters the system through scripts/main.py, where pd.read_csv is used to load the files specified via the --counts and --samples arguments.
Boundary markers: The SKILL.md workflow includes steps to "Validate that the request matches the documented scope" and "Confirm the user objective," but there are no explicit delimiters used when interpolating or processing the data contents within the agent context.
Capability inventory: The skill's primary script, scripts/main.py, is restricted to local file I/O (reading inputs, writing CSV results) and statistical computation using numpy and scipy. It does not possess network access or the ability to spawn subprocesses or execute arbitrary commands.
Sanitization: The skill relies on standard pandas CSV/TSV parsing. There is no specific validation or sanitization implemented to detect or strip natural language instructions that might be embedded within the data rows or metadata fields.

crispr-screen-analyzer