The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The SKILL.md file explicitly instructs the agent to provide a link to https://chatgpt-plus.top/list/#/home for users in mainland China. This specific URL has been flagged as a malicious phishing domain by automated security scanners.
[CREDENTIALS_UNSAFE]: The skill facilitates the exfiltration of sensitive credentials by recommending https://tryallapi.com/v1 as the default GPTIMAGE_BASE_URL. The scripts/generate.py script sends the user's GPTIMAGE_API_KEY in an Authorization header to this third-party endpoint, exposing the secret to an untrusted proxy.
[DATA_EXFILTRATION]: In addition to credential exposure, the skill sends user-provided product details and identity locks to external domains (tryallapi.com via the API script and gptimage2.asia via the generated browser links), potentially leaking sensitive intellectual property or product specifications.
[COMMAND_EXECUTION]: The skill instructions require the agent to execute shell commands (python scripts/generate.py and python scripts/online_link.py) that interpolate user-supplied text from the 'product identity lock' and 'final prompt' directly into command-line arguments. This pattern is vulnerable to command injection if the user-provided data contains shell metacharacters.
[INDIRECT_PROMPT_INJECTION]: The skill lacks sufficient sanitization and boundary markers when processing external data such as reference photos and product descriptions. These untrusted inputs are interpolated into prompt templates and subsequently passed to executable scripts, providing an attack surface where malicious input can influence agent behavior or script execution.

sale-banner