The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The PostDetail component in examples/fragments.md utilizes dangerouslySetInnerHTML to render the post.content field directly from a GraphQL query. This creates a vulnerability surface for Cross-Site Scripting (XSS) if the content returned by the server is malicious or compromised.
Ingestion points: External data from GraphQL query results (e.g., data.post.content in PostDetail).
Boundary markers: There are no explicit delimiters or instructions to the agent to treat this content as untrusted.
Capability inventory: While the code primarily handles rendering, the presence of an XSS vector combined with the skill's use of sensitive data in localStorage (auth tokens) increases the risk of data theft.
Sanitization: The provided examples do not include any sanitization or filtering logic (such as DOMPurify) before rendering raw HTML.
[SAFE]: The skill demonstrates standard and secure authentication patterns, including the dynamic attachment of Bearer tokens via Apollo's setContext link and the appropriate removal of tokens from localStorage upon authentication failure.
[SAFE]: The configuration for WebSocket-based subscriptions using graphql-ws follows architectural best practices, such as ensuring connections are only initiated in a browser environment and using functions to retrieve tokens at connection time.
[SAFE]: The use of official migration tools and code generation utilities, such as npx @apollo/client-codemod-migrate-3-to-4 and @graphql-codegen/cli, represents standard and expected development workflows.

web-data-fetching-graphql-apollo