The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions and README recommend installing a dependency skill using the command npx skills add modelstudioai/skills --skill bailian-docs-llm-wiki -y. As this targets the same vendor ('modelstudioai'), it is considered a standard operational dependency installation and does not pose an external threat.
[COMMAND_EXECUTION]: The skill requires the agent to execute local JavaScript files using Node.js (e.g., node scripts/recommend.mjs). These scripts perform deterministic logic such as filtering, scoring, and ranking models based on a local data catalog. No unsafe execution of arbitrary user-supplied code or shell commands was detected.
[DATA_EXPOSURE_AND_EXFILTRATION]: The scripts read model data from a local directory, typically within the user's home folder (~/.bailian/skills/...). The scripts are documented as 'pure I/O + pure functions' and do not perform network operations themselves. Code samples provided in the output use environment variables for sensitive API keys (e.g., os.environ["DASHSCOPE_API_KEY"]), following security best practices.
[INDIRECT_PROMPT_INJECTION]: The skill includes an attack surface where it processes user requirements to generate recommendations. However, the logic is handled by deterministic scripts that output structured JSON, and the instructions for the agent include explicit constraints on how to use that data, minimizing the risk of the model catalog content influencing the agent's behavior beyond the intended scope.

bailian-model-recommend