The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill uses curl to fetch markdown content from alphaxiv.org (in references/fetching-papers.md). This involves network operations to a non-whitelisted domain to retrieve external data based on user input.
[INDIRECT_PROMPT_INJECTION]: The skill is vulnerable to instructions embedded in the external papers it fetches, which could influence the agent's notebook planning and code generation stages.
Ingestion points: Research paper overviews and full texts are retrieved from alphaxiv.org via curl (as described in references/fetching-papers.md).
Boundary markers: There are no boundary markers or instructions directing the agent to ignore potential commands embedded within the retrieved research paper text.
Capability inventory: The agent can generate and execute Python code in marimo notebooks, create JavaScript components via anywidget, and perform file system operations using pathlib (referenced in references/anywidget.md).
Sanitization: The skill does not implement sanitization or validation of the paper content before using it to generate the notebook's structure and executable code.
[DYNAMIC_EXECUTION]: The skill involves dynamic code generation where the agent writes Python notebooks and JavaScript widgets based on external descriptions. The anywidget reference specifically encourages embedding JavaScript in Python strings and supports loading code from external paths via pathlib.

implement-paper