The Agent Skills Directory

[CREDENTIALS_UNSAFE]: The skill requires users to store sensitive credentials in plaintext configuration files or environment variables, including PayPal email and passwords, Cloudflare API tokens, and proxy authentication details.
[REMOTE_CODE_EXECUTION]: The installation process involves cloning a repository from an unverified source (https://github.com/DanOps-1/Gpt-Agreement-Payment) and executing its contents (pipeline.py), which orchestrates complex payment and browser automation flows.
[EXTERNAL_DOWNLOADS]: The skill downloads external resources including the main toolkit repository, browser engines via Playwright and Camoufox, and various ML models for CAPTCHA solving.
[COMMAND_EXECUTION]: Instructs the user to execute system-level commands including sudo apt-get install for the gost tunneling utility and playwright install for browser binaries.
[DATA_EXFILTRATION]: The skill's primary output is a collection of authentication refresh_token strings saved to output/results.jsonl, and it provides a local API endpoint that exposes system metadata and activity logs.
[PRIVILEGE_ESCALATION]: Requires the use of sudo to install system packages and networking tools like gost.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: Ingests external hCaptcha challenge data (images and prompts) from target websites via CTF-pay/hcaptcha_auto_solver.py.
Boundary markers: No delimiters or warnings are present in CTF-pay/hcaptcha_auto_solver.py to prevent malicious instructions within CAPTCHA prompts from influencing the VLM decision logic.
Capability inventory: The skill possesses extensive capabilities including network communication (requests, playwright), file system modifications (output/results.jsonl), and process execution (xvfb-run) across pipeline.py and CTF-pay/card.py.
Sanitization: No validation or sanitization of external challenge data is documented before processing.

gpt-agreement-payment-replay