The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill's installation instructions in SKILL.md and references/getting_started.md require users to install datacommons-client. However, the official Python client for Data Commons (linked in the skill's own documentation) is the datacommons package. Providing instructions to install an unofficial package name that imitates a well-known library is a characteristic of supply chain attacks, specifically typosquatting or dependency confusion.
[METADATA_POISONING]: The skill uses the name 'datacommons-client' and provides descriptions and links that point to official Data Commons documentation (docs.datacommons.org) and source code (github.com/datacommonsorg/api-python). This use of legitimate external references to authorize a different, potentially malicious package is deceptive.
[INDIRECT_PROMPT_INJECTION]: The skill processes statistical data and entity identifiers from an external API, which could be exploited to inject instructions if the source data is compromised or maliciously crafted.
Ingestion points: Data retrieved from the Data Commons knowledge graph through the Observation, Node, and Resolve endpoints.
Boundary markers: There are no markers or system instructions to ignore potential commands embedded in the statistical data or entity names.
Capability inventory: The skill possesses the capability to write data to the local file system (e.g., using pandas.DataFrame.to_csv) and output information to the console.
Sanitization: The provided examples do not demonstrate any validation or sanitization of the data received from the API before it is utilized in file operations or displayed to the user.

datacommons-client