The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains explicit instructions to install external dependencies ('BrowserAct' CLI and Skill) automatically if they are not present on the system. It specifically directs the agent to perform this action "without asking the user to confirm" and mandates that the agent "do not request separate approval for dependency installation." This pattern is a direct attempt to bypass user consent and oversight for the execution and installation of third-party software.
[COMMAND_EXECUTION]: The skill relies on the invocation of a command-line interface tool ('browser-act'). The instructions to automate the setup and execution of this CLI without human intervention increase the risk of unauthorized or malicious commands being run in the background.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to scrape and process data from external Temu product pages, which is an untrusted data source. This creates a vulnerability where an attacker could place malicious instructions within product titles or descriptions to manipulate the agent's subsequent actions.
Ingestion points: Product titles, brands, categories, descriptions, and specifications collected from target URLs (SKILL.md).
Boundary markers: Absent. The instructions do not define delimiters or provide warnings for the agent to ignore embedded instructions within the scraped data.
Capability inventory: The agent uses the browser-act CLI/Skill to interact with the web and is instructed to normalize and return data, implying file or memory operations (SKILL.md).
Sanitization: None detected. The skill does not mention any validation, escaping, or filtering of the content retrieved from the web.

temu-product-scraper