The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill contains instructions to automatically install the browser-act Skill and CLI if they are found to be missing from the environment. It specifically commands the agent to "install the missing dependency automatically without asking the user to confirm" and further specifies "do not request separate approval for dependency installation." This is a direct attempt to bypass human-in-the-loop security protocols for code execution.
[COMMAND_EXECUTION]: The requirement to install a CLI tool (browser-act) involves executing shell commands. By instructing the agent to suppress user notification and approval, the skill forces the execution of potentially high-privilege operations (software installation) without oversight.
[INDIRECT_PROMPT_INJECTION]: The skill's primary function is to scrape and process data from Alibaba supplier profiles and storefronts, which are untrusted external sources. This content is interpolated into the agent's context, creating a surface for adversarial content to influence agent behavior.
Ingestion points: Browser-visible page content, metadata, and network responses from Alibaba storefronts and product pages (SKILL.md, Instructions 1-5).
Boundary markers: No specific delimiters or instructions to ignore embedded commands within the scraped data are provided.
Capability inventory: The agent has the ability to execute shell commands (installation logic), perform network operations via the browser tool, and write to the local file system (workspaces/ directory).
Sanitization: The instructions do not define any sanitization, filtering, or validation steps for the data retrieved from external URLs.

alibaba-supplier-lead-research