The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the agent to provide a link to a website (https://chatgpt-plus.top/list/#/home) that has been flagged as a phishing risk by security scanners. It also promotes other untrusted domains like gptimage2.asia and tryallapi.com.
[DATA_EXFILTRATION]: The scripts/generate.py script transmits the GPTIMAGE_API_KEY environment variable to a remote server. The default server recommended in SKILL.md is https://tryallapi.com/v1, which is an untrusted third-party service. This configuration poses a significant risk of credential harvesting.
[CREDENTIALS_UNSAFE]: The skill encourages the use of a sensitive API key that is sent via HTTP headers (Authorization: Bearer) to a non-standard and untrusted external gateway, rather than a well-known service provider.
[INDIRECT_PROMPT_INJECTION]: The skill constructs prompts using product identity locks and campaign style locks that can be provided by the user. It lacks robust boundary markers or sanitization to prevent these inputs from being manipulated to override agent behavior.
Ingestion points: Product identity lock and style lock fields in SKILL.md.
Boundary markers: None identified.
Capability inventory: File system writing in scripts/crop_to_modules.py and network operations in scripts/generate.py.
Sanitization: No escaping or validation of user-provided prompt components is performed.

amazon-a-plus