The Agent Skills Directory

[PROMPT_INJECTION]: The skill exhibits an indirect prompt injection surface (Category 8) by processing untrusted data from Amazon product reviews.
Ingestion points: Untrusted review content is retrieved via the Sorftime API and stored in the reporting directory.
Boundary markers: The skill does not provide clear delimiters or 'ignore embedded instructions' warnings for the review analysis process.
Capability inventory: The agent possesses significant capabilities, including local file read/write operations and shell command execution via curl.
Sanitization: No explicit sanitization or filtering of the review text is mentioned prior to the agent's depth analysis.
[DATA_EXFILTRATION]: The skill retrieves an API key from a specific local path (D:/amazon-mcp/.mcp.json) and incorporates it into the query string of a curl request to mcp.sorftime.com. While this is functional, exposing secrets in a URL query parameter is a suboptimal security practice as it may lead to credential exposure in server logs, proxy logs, or shell history.
[COMMAND_EXECUTION]: The skill relies on shell commands (curl, mkdir, cp) to interact with external services and manage local files. It interpolates user-supplied data, such as the product ASIN and site identifier, directly into these commands and JSON payloads. This presents a potential risk of injection if the agent does not strictly validate or sanitize these inputs before execution.

review-analysis