figma-canon
Pass
Audited by Gen Agent Trust Hub on Oct 8, 2026
Risk Level: SAFECOMMAND_EXECUTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFE
Full Analysis
- [COMMAND_EXECUTION]: The skill provides detailed instructions for the agent to execute shell commands, including
npm install,node,python3,kill,lsof, andcurl. Notably,references/field-notes.mddescribes a recovery mechanism that involves spawning a background daemon usingnohup node <mcp-direct>/daemon.mjs &to bypass connectivity issues. - [DYNAMIC_EXECUTION]: The skill's primary functionality relies on generating and executing JavaScript code within the Figma plugin sandbox using the
figma_executetool. This allows the agent to perform complex design and data operations, but represents a significant capability that must be handled carefully. - [EXTERNAL_DOWNLOADS]: The skill directs the agent to download and install Node.js packages like
@figma/code-connectand fetch assets from external sources such asitunes.apple.comfor icon retrieval. These sources are generally well-known, but represents the ingestion of external code and data. - [INDIRECT_PROMPT_INJECTION]: The skill is designed to process external data from Figma files (layer names, text content, annotations), which the skill explicitly identifies in
references/security-canon.mdas untrusted sources that could potentially harbor prompt injection attacks. - [CREDENTIALS_UNSAFE]: The skill mentions the use of the
FIGMA_ACCESS_TOKENenvironment variable for authentication. It provides safety instructions for handling this token, such as not printing it to logs or committing it to repositories, but acknowledges the presence of sensitive credentials in the environment.
Audit Metadata