The Agent Skills Directory

[METADATA_POISONING]: The skill file contains significantly inconsistent documentation sections. While the primary instructions describe a simple link generator for a third-party website service, the subsequent 'Overview', 'Prerequisites', 'Instructions', 'Output', and 'Error Handling' sections describe a complex system-level tool capable of modifying project configuration files, using CLI tools, and interacting with APIs. This contradictory metadata misrepresents the skill's actual capabilities and execution environment, as the allowed-tools configuration is strictly limited to AskUserQuestion.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest untrusted data from users to construct functional URLs, creating an attack surface for injection.
Ingestion points: Business names and business descriptions provided by the user in the chat environment.
Boundary markers: Absent; there are no delimiters used to isolate user-provided strings from the link template.
Capability inventory: The skill is restricted to AskUserQuestion, though the metadata claims capabilities for file modification and API interaction.
Sanitization: The instructions explicitly require URL-escaping of the business name and description before interpolation into the B12 signup link, which is a required safety practice.

website-generator