The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external content through the aeo_audit.py script, which presents a surface for indirect prompt injection attacks.
Ingestion points: The scripts/aeo_audit.py script fetches content from user-provided URLs using urllib.request.urlopen (File: scripts/aeo_audit.py).
Boundary markers: Absent. While the skill instructs the agent on AEO methodology, it lacks explicit instructions to treat the audited HTML content as untrusted or to ignore any natural language instructions found within that content.
Capability inventory: The skill possesses network-read capabilities (via audit script) and file-write capabilities (via prompt matrix script).
Sanitization: The script utilizes html.parser.HTMLParser to extract text and JSON-LD, but does not implement filtering or sanitization to prevent the agent from interpreting instructions embedded in the target site's HTML.
[EXTERNAL_DOWNLOADS]: The scripts/aeo_audit.py script performs network requests to fetch external resources for auditing purposes.
The script uses Python's standard urllib.request library to download HTML content from arbitrary URLs provided to it at runtime.
[COMMAND_EXECUTION]: The skill requires the execution of bundled Python scripts to perform its core functions.
The agent is instructed to run python3 scripts/aeo_audit.py and python3 scripts/build_prompt_matrix.py to audit sites and generate experimentation sets. These scripts are self-contained and do not rely on external non-standard dependencies.

aeo