The Agent Skills Directory

[DYNAMIC_EXECUTION]: The script sci_search.py contains an _init_shared_env function that uses importlib.util to dynamically load and execute a Python module from a relative path (../_shared/env_config.py). Loading and executing code from computed paths at runtime is a risk factor as it can be exploited if an attacker can write to sibling directories.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external academic APIs (arXiv, PubMed, WoS, Springer, Scopus, Semantic Scholar, OpenAlex).
Ingestion points: Data is fetched in the search methods of various fetcher classes in sci_search.py.
Boundary markers: There are no explicit delimiters or instructions to ignore embedded commands within the fetched paper titles or abstracts.
Capability inventory: The skill has network access (urllib.request) and file writing capabilities (PaperLibrary cache).
Sanitization: The script performs basic string cleaning but does not sanitize potential prompt injection patterns within the retrieved paper content.
[COMMAND_EXECUTION]: The SKILL.md documentation provides examples of executing the Python script via the command line with user-controlled arguments, which are then parsed and used to construct API queries.

sci-search