The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes data from Alibaba Cloud SLS logs and Datasets. This data is untrusted and is interpolated into LLM prompts within Pipeline nodes (e.g., llm-call and agentic-call). The skill mitigates this risk by using structured templates and providing clear documentation on data handling, though the ingestion of external data remains a vulnerability surface.
[EXTERNAL_DOWNLOADS]: Documentation includes instructions for downloading and executing scripts for tool installation (Aliyun CLI) and framework-specific observability plugins (OpenClaw, CoPaw, Hermes). Downloads for the Aliyun CLI originate from verified vendor domains. For framework-specific tools, the skill includes explicit security warnings advising users to review script contents before execution rather than piping directly to a shell.
[COMMAND_EXECUTION]: Bundled Python scripts (agentloop_eval.py, agentloop_pipeline.py) utilize subprocess.run to execute the aliyun CLI. These operations are strictly controlled via a binary allow-list and argument validation that prevents null bytes or newline character injection.

alibabacloud-agentloop-management