The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from external sources, which creates a surface for indirect prompt injection.
Ingestion points: Research sub-agents gather literature from the web and save them to /mnt/agents/output/research/. Additionally, the artifact_analyst processes user-supplied materials from /mnt/agents/upload/.
Boundary markers: The skill uses structured templates (e.g., ## Input Materials in content.md) to delimit external content, which provides some protection but does not eliminate the risk of the agent obeying instructions embedded in the source text.
Capability inventory: The system can create sub-agents, write files to the local filesystem, and execute Python code via IPython for visualization.
Sanitization: No explicit sanitization, escaping, or validation of the ingested external content is mentioned before it is interpolated into sub-agent prompts.
[COMMAND_EXECUTION]: The skill instructs writer sub-agents to use the IPython tool to execute matplotlib and seaborn code for generating figures and charts. While this is standard for data visualization, it involves the execution of dynamically generated code.
[EXTERNAL_DOWNLOADS]: The workflow relies on research sub-agents performing web searches to acquire literature dimensions and citations. While these are performed through presumably standard search tools, they introduce external content into the agent's context.

paper-writing