The Agent Skills Directory

[DATA_EXPOSURE]: The skill is configured to access sensitive local browser history database files. Specifically, it instructs the agent to read from paths such as ~/Library/Application Support/Google/Chrome/Default/History. These files contain a comprehensive record of a user's web activity, which is considered highly sensitive personal data.
[COMMAND_EXECUTION]: The skill requires the sqlite3 command-line utility and provides procedures for the agent to execute shell commands. This includes copying the local Chrome history database to a temporary location and running SQL queries against it to extract search terms and listing visits.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data sourced from external website titles and URLs stored in the browsing history, creating a potential surface for indirect injection attacks.
- Ingestion points: Chrome history SQLite database (History file).
- Boundary markers: Absent. The skill does not define specific delimiters or instructions to ignore potential command-like text within the ingested history data.
- Capability inventory: Terminal access for executing sqlite3 queries and file system access for copying and reading database files.
- Sanitization: While the skill includes logic to normalize phrases and filter them against a user brief, it lacks specific sanitization or filtering to detect or remove prompt injection payloads that might be embedded in page titles or URL parameters.

facebook-marketplace-history-seed