Docs Canvas — agentic threat model

Not certain from the listing — relies on Cursor's underlying foundation models. Primary threats include prompt injection via malicious codebase comments or documentation files designed to hijack the rendering output or generate misleading diagrams.

Reads local repository documentation, API references, and source code. Threat of data exfiltration if the model or plugin is compromised and transmits sensitive intellectual property or hardcoded secrets from the codebase to external endpoints.

Orchestrates the parsing of files to generate a structured Canvas view. Risk of insecure tool integration if the parsing logic or diagram generation tools are vulnerable to path traversal or command injection via malformed source files.

Not certain from the listing — runs locally as a Cursor plugin. Security posture depends entirely on the host IDE's sandboxing, local file permissions, and network isolation capabilities.

Not certain from the listing — likely lacks dedicated observability or guardrails, relying on the user to visually inspect the generated Canvas and diagrams for accuracy and safety.

Not certain from the listing — lacks explicit mention of compliance, access controls, or audit logging. Relies on the user's local workstation security and repository access permissions.

Operates as an isolated plugin within the IDE ecosystem. Minimal multi-agent risk, though it could theoretically interact with other installed Cursor plugins or extensions if they share the same workspace context.