The Agent Skills Directory

[DATA_EXFILTRATION]: The aeo_audit.py script performs network requests to external URLs provided by the user using the urllib.request library. While this is a documented feature for auditing live websites, it represents a network operation to non-whitelisted domains.
Evidence: The fetch_url function in scripts/aeo_audit.py uses urllib.request.urlopen to retrieve HTML content from user-specified URLs.
[PROMPT_INJECTION]: The skill ingests and processes untrusted data from external websites and local files, creating a potential vector for indirect prompt injection.
Ingestion points: Untrusted content enters the agent's context through URL fetching in scripts/aeo_audit.py and file reading in scripts/aeo_audit.py and scripts/aeo_optimizer.py.
Boundary markers: The ingested content is not isolated with protective delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has the capability to perform network reads (scripts/aeo_audit.py), write to the local file system (scripts/citation_tracker.py), and present processed content to the agent for further action.
Sanitization: There is no evidence of sanitization or filtering to detect or neutralize malicious prompt injection instructions within the ingested content.

aeo