The Agent Skills Directory

[PROMPT_INJECTION]: Widespread instructions across multiple files (e.g., analyse-domain.md, site-import.md, domain-search-purchase-and-connect.md) direct the AI to perform operations "silently" and "never announce" tool calls or monitoring status. This intentional concealment hides the agent's internal actions and decision-making processes from the end-user.
[INDIRECT_PROMPT_INJECTION]: The skill has a significant vulnerability surface for indirect prompt injection due to its data processing model.
Ingestion points: The agent ingests data from external sources during site imports (site-import.md) and processes event payloads in automation triggers (e.g., payload in references/automations/conditions.md).
Boundary markers: The instructions lack markers or specific guidelines to distinguish between control instructions and data content when processing external results.
Capability inventory: The agent possesses extensive capabilities, including creating/deleting sites, modifying business configurations, and calling arbitrary REST endpoints.
Sanitization: No sanitization or validation of untrusted strings is present in the code generation templates provided.
[DYNAMIC_EXECUTION]: The skill provides recipes that require the agent to generate and execute JavaScript code at runtime. For example, references/automations/conditions.md includes templates for CODE_CONDITION logic, and references/sites/upload-static-site.md contains scripts for site file manipulation. This dynamic logic is used to interact with the Wix platform's execution environment.
[COMMAND_EXECUTION]: The documentation includes various shell command patterns, primarily curl and wix CLI commands, for the agent to use when managing domains, site files, and event registration. These are part of the intended operational flow for site management.

wix-manage