The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the creation of bioinformatics pipelines that ingest untrusted data from external files and database records, creating a surface for indirect prompt injection.
Ingestion points: The skill uses LatchFile and LatchDir in references/data-management.md and SKILL.md to process external sequencing data and Registry records.
Boundary markers: There are no explicit instructions or delimiters provided to the agent to distinguish between legitimate data and potential instructions embedded within the processed biological files.
Capability inventory: The skill requires Bash(python:*) and Write access to generate and execute workflow code, which includes network operations (via latch login and latch register) and file system access.
Sanitization: No evidence of data validation or sanitization was found in the provided code templates or instructions before data is used in workflow tasks.

alterlab-latchbio