The Agent Skills Directory

[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it processes content from user-provided files and interpolates it into prompts for the Anthropic API without sufficient sanitization.\n
Ingestion points: The autoresearch.py script reads text from the file path passed to the --input argument.\n
Boundary markers: While the script uses --- delimiters in prompts to separate instructions from the content, it does not provide explicit directives to the model to ignore instructions embedded within the content.\n
Capability inventory: The skill possesses capabilities to read local files, write new files to the data/ directory, and perform network requests to the Anthropic API.\n
Sanitization: No sanitization or filtering logic was identified for the input content prior to its inclusion in the model prompts.\n- [COMMAND_EXECUTION]: The skill requires the execution of the autoresearch.py script to perform its core functions. The script interacts with the local file system and network, which is expected for its stated purpose.\n- [EXTERNAL_DOWNLOADS]: The skill specifies the anthropic library in its requirements.txt file, which is downloaded from the official Python Package Index (PyPI) during installation. This is a standard and safe dependency.\n- [DATA_EXFILTRATION]: The skill sends content from the input files to the Anthropic API (api.anthropic.com) for processing. This network communication is a necessary component of the skill's optimization workflow.

autoresearch