The Agent Skills Directory

[COMMAND_EXECUTION]: The skill facilitates the execution of shell commands and Python code, primarily through the get_results() stage. This stage uses an automated 'engineer' agent to run computational experiments, perform statistical analysis, and generate visualizations.
[DYNAMIC_EXECUTION]: The skill employs dynamic code generation and execution. The agent-driven workflow translates research methodologies into executable Python code at runtime to process data and produce research findings.
[INDIRECT_PROMPT_INJECTION]: The skill's architecture is susceptible to indirect prompt injection as it ingests untrusted data descriptions and fetches literature from external APIs like Semantic Scholar and Perplexity.
Ingestion points: Dataset descriptions provided via set_data_description() in SKILL.md and literature search results retrieved via check_idea() in references/research_pipeline.md.
Boundary markers: The documentation does not specify explicit delimiters or isolation mechanisms to prevent embedded instructions in research data from influencing the agent's logic.
Capability inventory: The skill possesses extensive capabilities, including system command execution (Bash), network access for literature searches, and file system writes for project artifacts.
Sanitization: No explicit sanitization or filtering of input data is described before it is processed by the orchestration agents.
[EXTERNAL_DOWNLOADS]: The installation process involves downloading software from external sources, including the denario Python package via uv or pip, Docker images from Docker Hub, and the Google Cloud SDK. The documentation correctly identifies the risks associated with unverified remote script execution and recommends using package managers instead.

alterlab-denario