claude-design-premium-harness
Fail
Audited by Gen Agent Trust Hub on Jun 18, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill's installation steps include downloading a ZIP archive from a personal GitHub repository at
https://github.com/oalanicolas/claude-design-premium/archive/refs/heads/main.zip. This source is not verified as a trusted vendor. - [REMOTE_CODE_EXECUTION]: Users are instructed to run downloaded JavaScript files using the
nodecommand (for example,node scripts/bootstrap-harness.mjs). This combined with the external download source constitutes a remote code execution risk. - [COMMAND_EXECUTION]: Documentation provides shell commands like
curl,unzip, andnodeto be executed by the user, facilitating the retrieval and execution of external code. - [PROMPT_INJECTION]: The skill ingests untrusted external data into the agent's context, creating an indirect prompt injection vulnerability.
- Ingestion points: Project configuration files such as
DESIGN.md,_ds_manifest.json, andBOUND_DS.jsonare read and processed. - Boundary markers: None identified. There are no delimiters or instructions to ignore commands within the design system files.
- Capability inventory: The skill performs file system operations and executes Node.js scripts.
- Sanitization: No sanitization or validation logic is present for the content of the design system manifest or brief files.
Recommendations
- AI detected serious security threats
Audit Metadata