The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/search_flights.py utilizes the AMEX_2FA_COMMAND environment variable to execute user-defined shell commands via subprocess.run(shell=True). This mechanism is designed for 2FA automation but allows for arbitrary command execution in the host environment.
[CREDENTIALS_UNSAFE]: The skill requires users to provide their American Express credentials through environment variables (AMEX_USERNAME, AMEX_PASSWORD). Additionally, it persists sensitive session information by saving browser cookies to a local file (cookies.json) in the user's profile directory.
[EXTERNAL_DOWNLOADS]: The skill depends on patchright, which is identified as an 'undetected Playwright fork'. It also references a custom Docker image ghcr.io/borski/amex-travel. These dependencies on non-standard or vendor-specific forks represent a potential supply chain risk.
[PROMPT_INJECTION]: The hotel search functionality scrapes hotel names, amenities, and benefit descriptions directly from the DOM using innerText and regex patterns. This external, untrusted content is then presented to the agent without sanitization, creating a surface for indirect prompt injection.
Ingestion points: DOM scraping in scripts/search_flights.py (specifically extract_app_data_hotels and _parse_offer_card_text).
Boundary markers: Absent. The scraped data is converted into markdown tables for agent consumption without delimiters.
Capability inventory: The skill has the ability to execute shell commands (subprocess.run) and perform broad network operations through the automated browser.
Sanitization: None. The script performs raw text extraction and regex-based parsing from the browser context.

amex-travel