The Agent Skills Directory

[SAFE]: The skill provides static documentation and reference guides for the LLaMA-Factory project. No malicious content or instructions were identified during the analysis of the skill body or reference files.
[EXTERNAL_DOWNLOADS]: Reference files such as references/advanced.md and references/getting_started.md contain links to official project documentation and images hosted on llamafactory.readthedocs.io. These are recognized as well-known and trusted sources for developer documentation.
[INDIRECT_PROMPT_INJECTION]: The skill has an inherent attack surface for indirect prompt injection because it processes documentation and is designed to assist with user-supplied data (such as datasets or logs) related to machine learning tasks.
Ingestion points: Reference files in the references/ directory and user-provided queries, code, or error logs used during debugging sessions.
Boundary markers: No specific delimiters or warnings for the agent to ignore embedded instructions are present in the documentation.
Capability inventory: The skill does not include any executable scripts or tools. It lacks capabilities such as file-writing, network exfiltration, or command execution, which limits the potential impact of an injection.
Sanitization: No explicit sanitization or filtering of external content is implemented within the provided reference materials.

llama-factory