idea-to-canvas
Pass
Audited by Gen Agent Trust Hub on Aug 23, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection. It is explicitly designed to ingest untrusted data sources such as 'Slack dumps', 'meeting notes', and 'design docs' and process them into a structured output. A malicious actor could embed instructions within these source materials to influence the agent's behavior during the transformation process.
- Ingestion points: The
@<input>variable inSKILL.mdaccepts arbitrary user-provided text. - Boundary markers: None. The instructions do not include delimiters or specific guidance to the model to ignore instructions embedded within the source text.
- Capability inventory: The skill uses the
Cursor Canvastool to generate.canvas.tsxfiles. While it does not perform direct shell execution, the generation of executable UI code (TSX) from untrusted input increases the potential impact of an injection. - Sanitization: None observed. The instructions focus on preserving intent and terminology without mentioning validation or sanitization of the input content.
Audit Metadata