The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external academic papers, which could contain malicious instructions designed to subvert agent behavior.
Ingestion points: Data is ingested via scripts/main.py using the --input, --text, and --url command-line arguments.
Boundary markers: The script does not implement delimiters or instructions for the agent to ignore directives embedded within the research paper content.
Capability inventory: The script has the capability to read local files, execute network GET requests, and write output to the file system.
Sanitization: No validation or sanitization of the input text is performed before it is processed by regular expressions and scored.
[DATA_EXFILTRATION]: The scripts/main.py implementation uses the requests library to perform network GET operations for fetching paper content from external, non-whitelisted URLs.
[COMMAND_EXECUTION]: The skill workflow requires the execution of the Python script scripts/main.py, which handles sensitive operations like file system access and network communications.
[METADATA_POISONING]: The repository includes a file abstract-summarizer_audit_result_v2.json that contains extensive, unverified claims regarding the skill's safety and effectiveness. These claims are treated as deceptive metadata and are disregarded in the final safety evaluation.

abstract-summarizer