The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/search_flights.py executes shell commands defined in an environment variable using shell=True, which can lead to arbitrary code execution if the variable is manipulated.
Evidence: The script retrieves a command from AMEX_2FA_COMMAND and executes it via subprocess.run(hook_cmd, shell=True, ...) to automate 2FA code retrieval.
[CREDENTIALS_UNSAFE]: The skill manages sensitive financial account credentials and session tokens.
Evidence: It requires AMEX_USERNAME and AMEX_PASSWORD environment variables and persists browser session cookies to ~/.amex-travel-profiles/cookies.json to bypass future 2FA prompts.
[INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect injection as it processes untrusted data from the Amex travel portal and user-supplied search strings.
Ingestion points: User inputs for origins, destinations, and dates; external flight and hotel data extracted from the American Express travel portal.
Boundary markers: The skill does not implement explicit delimiters or instruction-ignore warnings when interpolating extracted data into its output.
Capability inventory: The skill has browser-based network access, file system write access for session/status tracking, and shell command execution capabilities.
Sanitization: While there is basic validation for dates and cabin types, search parameters are directly typed into the browser automation tool.

amex-travel