The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill instructions attempt to read an API key from a hardcoded sensitive file path at /Users/lingzhi/Code/keys.md.
The extracted key is passed directly as a command-line argument (--api-key), which can expose the secret in the system's process table (e.g., via ps commands) to other local users.
[COMMAND_EXECUTION]: The skill uses shell command substitution ($(grep ... )) within SKILL.md to dynamically inject secrets into execution context.
Multiple Python scripts are invoked via the shell, relying on the environment to manage execution.
[INDIRECT_PROMPT_INJECTION]: The skill accepts natural language search queries from the user ($ARGUMENTS) and interpolates them into shell commands, creating a potential surface for injection attacks.
Ingestion points: SKILL.md accepts search queries from the user.
Capability inventory: The skill has the capability to execute shell commands and write to the local filesystem.
Boundary markers: No explicit delimiters or boundary markers are used to separate user input from command instructions.
Sanitization: The skill does not perform explicit sanitization or validation of the search query before it is passed to script arguments.
[EXTERNAL_DOWNLOADS]: The skill downloads academic metadata and source code from well-known academic services.
It fetches paper details from the arXiv API and downloads source tarballs from arxiv.org.
It queries the CrossRef API (api.crossref.org) and the OpenAlex API (api.openalex.org) for bibliographic data.

literature-search