The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data in multimodal image-text search and knowledge base search scenarios. It ingests content from CSV, JSON, and TXT files to generate embeddings and populate search indexes. There is no evidence of boundary markers or sanitization logic to handle potentially malicious instructions embedded within these data sources, creating a risk that attacker-controlled data could influence agent behavior.
Ingestion points: CSV product lists and knowledge documents (txt/json) referenced in references/01-dev/multimodal-search-scene.md and references/01-dev/knowledge-search-scene.md.
Boundary markers: Absent in instructions for data ingestion.
Capability inventory: High-privilege CLI execution (aliyun, lindorm-cli) and network operations (curl).
Sanitization: Not specified for external data content.
[EXTERNAL_DOWNLOADS]: Fetches official database management tools and CLI binaries from well-known Alibaba Cloud Object Storage Service (OSS) buckets and content delivery networks. This includes aliyuncli.alicdn.com, tsdbtools.oss-cn-hangzhou.aliyuncs.com, and hbaseuepublic.oss-cn-beijing.aliyuncs.com.
[REMOTE_CODE_EXECUTION]: Provides automated shell scripts and manual instructions to download, extract, and execute binary packages for lindorm-cli and alihbase shell. These patterns are used for installing official vendor-provided tooling.
[PRIVILEGE_ESCALATION]: Directs the user to use sudo for moving downloaded binaries into restricted system directories such as /usr/local/bin/ during the Aliyun CLI installation process.

alibabacloud-lindorm-agent-skill