The Agent Skills Directory

[Indirect Prompt Injection] (SAFE): The skill analyzes user-provided workflow code (.nf, .smk, .config), which presents a surface for indirect prompt injection if those files contain malicious instructions intended for the agent.
Ingestion points: User-provided bioinformatics workflow definitions targeted by the regex patterns in references/validations.md.
Boundary markers: The SKILL.md lacks explicit XML delimiters or system-level warnings to the agent to disregard instructions embedded within the processed code files.
Capability inventory: The skill is entirely descriptive and diagnostic; it does not include any executable scripts (.py, .js, .sh), and thus has no capability for command execution, file modification, or network access.
Sanitization: No sanitization or filtering logic is present for the ingested code content.
[Unverifiable Dependencies] (SAFE): No external package dependencies (Python/Node.js) or remote script downloads are used in this skill.
[Privilege Escalation] (SAFE): While the skill mentions container volume mounting (e.g., -v /data:/data), these are presented as educational examples for users rather than commands executed by the agent itself.

bioinformatics-workflows