The Agent Skills Directory

[PROMPT_INJECTION]: The scripts/parse_snapshot.py script processes external accessibility tree snapshots that are generated from webpages. If the source webpage contains malicious instructions, they could be reconstructed into the final Markdown and subsequently interpreted by the agent as commands. \n- Ingestion points: The script reads external data from the file path provided to the --input argument in scripts/parse_snapshot.py. \n- Boundary markers: The script does not include delimiters or warnings in its output to instruct the agent to ignore potentially embedded instructions within the reconstructed Markdown. \n- Capability inventory: The script's capabilities are limited to local file system read and write operations. It contains no network communication, subprocess execution, or dynamic code evaluation logic. \n- Sanitization: No sanitization, filtering, or escaping of the textual content is performed during the parsing process, which allows instructions in the source tree to be passed through to the agent context.

create-md-from-browsermcp-snapshot