The Agent Skills Directory

[DYNAMIC_EXECUTION]: The CodeDemo.vue component example uses eval(props.code) to execute JavaScript code passed via a component prop. This allows for dynamic code execution at runtime, which is a high-risk pattern if the source of the code prop is untrusted or maliciously crafted slide content.
[REMOTE_CODE_EXECUTION]: The inclusion of eval() in a component that accepts external props creates a potential Remote Code Execution (RCE) vector. If an attacker can control the input to the CodeDemo component (e.g., through a shared presentation or external data source), they can execute arbitrary commands in the context of the Slidev application.
[EXTERNAL_DOWNLOADS]: The skill includes examples that fetch data and scripts from external domains, such as the Tweet and Youtube components. It also references the qrcode Node.js package in the QRCode.vue example. While these are common presentation features, they represent connections to external third-party infrastructure.
[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for indirect prompt injection by creating components that ingest and process external parameters (code in CodeDemo, url in QRCode, id in Youtube) without explicit sanitization or boundary markers to prevent the execution of embedded instructions.
Ingestion points: CodeDemo.vue (code prop), QRCode.vue (url prop), Youtube (id prop).
Boundary markers: Absent; the components process props directly without "ignore instructions" delimiters.
Capability inventory: The skill includes eval() execution capabilities in the CodeDemo example and network-fetching capabilities via external media components.
Sanitization: No input validation or sanitization is present in the provided component examples.

slidev-components