The Agent Skills Directory

[COMMAND_EXECUTION]: The skill documents the implementation and usage of the commandline tool component (github.com/cloudwego/eino-ext/components/tool/commandline), which is designed to allow AI agents to execute shell commands in the host environment.
[INDIRECT_PROMPT_INJECTION]: The framework described in the skill features a document processing pipeline that ingests data from external, untrusted sources, creating a vulnerability surface for indirect prompt injection.
Ingestion points: Document loaders fetch data from HTTP URLs (url.NewLoader), AWS S3 buckets (s3.NewS3Loader), and the local file system (file.NewFileLoader), as detailed in reference/document/pipeline.md.
Boundary markers: The documentation does not demonstrate the use of delimiters or explicit instructions to the model to ignore embedded commands in the ingested data.
Capability inventory: The framework provides high-privilege tools including commandline (shell execution), httprequest (network access), and browseruse (browser automation), which could be targets for manipulation via injected instructions.
Sanitization: No specific sanitization or filtering logic for external data is shown in the provided configuration examples.
[EXTERNAL_DOWNLOADS]: The skill includes instructions to download framework extensions using go get from the author's GitHub repository. It also references the execution of remote packages via npx when configuring Model Context Protocol (MCP) servers (reference/tool/mcp.md).
[DYNAMIC_EXECUTION]: The skill supports dynamic content generation and logic evaluation through prompt templates using Go text/template and Jinja2 syntax (reference/prompt.md). Additionally, the InvokableTool interface executes logic based on dynamically generated JSON arguments.

eino-component