The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill instructs the agent to dynamically load and execute code from bundled helper scripts using the exec(open("<skills>/.../kernel.py").read()) pattern. This method bypasses standard module import safety mechanisms by reading and executing file content directly in the runtime environment.
[COMMAND_EXECUTION]: The instructions direct the agent to run various Python scripts throughout the research lifecycle, such as project_status.py for status checks and init_project.py for project setup. These are directives for arbitrary code execution within the agent's environment.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and analyzing untrusted external content, creating a vulnerability surface.
Ingestion points: Processing external research data through tools like search_openalex and extracting content from PDF documents using pdf-explore (outline, pages, scan, and grep functions).
Boundary markers: The skill does not specify any delimiters, quoting, or instructions to ignore potential instructions embedded within the research papers or metadata being analyzed.
Capability inventory: The agent has the capability to write to the project directory, execute local Python scripts via exec(), and run CLI tools for project management.
Sanitization: There is no mention of sanitizing, filtering, or validating the external content retrieved from academic databases or PDF files before it is processed by the agent's logic.

research-methodology