The Agent Skills Directory

[COMMAND_EXECUTION]: The skill executes shell commands including mkdir, cd, and git clone. It also runs a local Python script scripts/github_search.py to interface with the GitHub API.
[EXTERNAL_DOWNLOADS]: The skill performs network requests to GitHub's official API (api.github.com) to search for repositories. It subsequently clones entire repositories from external URLs provided by the search results. While GitHub is a well-known service, the specific content within the cloned repositories is untrusted and originates from various third-party authors.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection (Category 8) because it ingests and processes untrusted external data.
Ingestion points: Content from any repository cloned into the _examples/ directory, which the agent is instructed to read using tools like smart_read or grep_search.
Boundary markers: Absent. There are no instructions provided to wrap the external content in delimiters or to ignore embedded instructions found within the cloned code or README files.
Capability inventory: The skill possesses the ability to create directories, clone repositories, and execute a search script using the requests library. The agent environment likely includes additional tools for file reading and processing.
Sanitization: Absent. The skill does not perform any validation, filtering, or escaping of the content found in the third-party repositories before the agent inspects it.

find-examples