The Agent Skills Directory

[DATA_EXFILTRATION]: The script contains a function _read_zshrc that opens and parses the user's shell configuration file (~/.zshrc) to extract the ALPHAXIV_TOKEN. Reading shell profile files is a high-risk activity as these files often contain sensitive environment variables, aliases, and system credentials beyond the specific token requested.
[COMMAND_EXECUTION]: The _post_stream function in scripts/alphaxiv.py uses subprocess.run to execute the curl binary. This is used to handle text/event-stream responses for the ask command. Executing external binaries via subprocess is a sensitive operation that can lead to command injection if input sanitization is bypassed.
[PROMPT_INJECTION]: The skill is vulnerable to indirect prompt injection because it ingests and processes untrusted data from the AlphaXiv API, including paper abstracts, titles, and AI-generated overviews.
Ingestion points: Paper metadata and summaries are fetched via _get calls in scripts/alphaxiv.py and displayed to the agent.
Boundary markers: Absent. The content retrieved from the API is interpolated directly into the output sent back to the agent without delimiters or warnings to ignore embedded instructions.
Capability inventory: The skill has capabilities for file system access (reading ~/.zshrc) and subprocess execution (curl) in scripts/alphaxiv.py.
Sanitization: Absent. The script performs no validation or filtering of the text content returned by the API before presenting it to the agent.

alphaxiv