The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/check_balance.py uses subprocess.run(hook_cmd, shell=True) to execute a command provided through the AA_2FA_COMMAND environment variable. While intended as a feature for automating 2FA code retrieval, the use of shell=True on an externally-provided string allows for arbitrary shell command execution.
[CREDENTIALS_UNSAFE]: The skill implements a persistence mechanism by saving browser profiles, including session cookies, to ~/.aa-browser-profiles/. This allows future runs to bypass 2FA but stores sensitive authentication tokens in a predictable location on the file system.
[EXTERNAL_DOWNLOADS]: The skill requires the installation of patchright, an external library used for bypassing bot detection, and references Docker images hosted on ghcr.io/borski/. These are vendor-specific resources required for the skill's primary function.
[DATA_EXPOSURE]: The script saves debug screenshots to /tmp/aa-debug.png and reads 2FA codes from /tmp/aa-2fa-code.txt. In shared environments, sensitive account information or 2FA codes in `/tmp/ could be accessible to other local users.
[INDIRECT_PROMPT_INJECTION]: The skill has a surface for indirect prompt injection as it processes and extracts text from external web pages (AA.com).
Ingestion points: Data is extracted from the American Airlines account summary page in scripts/check_balance.py using page.evaluate().
Boundary markers: None. The script directly parses the page content.
Capability inventory: The script has subprocess.run capabilities (for 2FA hooks) and file system access (for profiles and temporary files).
Sanitization: Extracted data is validated against regex patterns or converted to integers, which provides some mitigation against malformed data.

american-airlines