korean-cinema-search
Fail
Audited by Gen Agent Trust Hub on Oct 1, 2026
Risk Level: HIGHEXTERNAL_DOWNLOADSREMOTE_CODE_EXECUTIONCOMMAND_EXECUTIONPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill downloads code and assets from unverified external sources, including the npm registry (packages '@nomadamas/k-skill' and 'daiso') and a third-party GitHub repository ('hmmhmmhm/daiso-mcp').
- [REMOTE_CODE_EXECUTION]: The skill implements multiple remote code execution patterns. It requires running 'npx -y @nomadamas/k-skill@0' to fetch instructions and 'npx --yes daiso' for theater lookups. It also provides a fallback method involving 'git clone', 'npm install', and executing built artifacts with 'node dist/bin.js'.
- [COMMAND_EXECUTION]: The skill's core functionality relies on executing shell commands, providing a significant capability for an attacker if the inputs or the downloaded tools are compromised.
- [PROMPT_INJECTION]: The skill contains instructions to prioritize dynamic content fetched via 'npx' as the 'primary instructions', which acts as a meta-instruction that could override safety guidelines or local security analysis.
- [INDIRECT_PROMPT_INJECTION]: The skill processes data from external cinema APIs and websites (cgv.co.kr), creating a vulnerability where malicious content in those sources could influence the agent's behavior.
- Ingestion points: Output from 'daiso' CLI and content from 'cgv.co.kr' accessed via browsing tools.
- Boundary markers: No delimiters or warnings are specified for the external content.
- Capability inventory: Shell command execution (npx, node, git), file system writes (npm install), and network operations.
- Sanitization: No sanitization or validation of the external output is mentioned.
Recommendations
- AI detected serious security threats
Audit Metadata