The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs developers to download the official Google Agent Development Kit using standard Go toolchain commands (e.g., go get google.golang.org/adk@v1.2.0). This targets a well-known, trusted organization namespace.
[COMMAND_EXECUTION]: Documentation includes example CLI commands for developers to execute their own local code, such as go run agent.go web api. These are standard development workflows for the ADK runtime.
[INDIRECT_PROMPT_INJECTION]: The skill architecture handles user-provided data. It mitigates injection risks by instructing developers to implement safety controls, including input/output sanitization, tool confirmations, and execution guardrails.
Ingestion points: Data enters via llmagent inputs and the Runner.Run interface defined in core-primitives.md.
Boundary markers: Systems are partitioned using explicit Instruction blocks and agent Description fields to guide model behavior.
Capability inventory: The skill provides a framework for custom tools, MCP integrations, and sub-agent delegation.
Sanitization: Developers are explicitly directed in tools-and-guardrails.md to use AfterToolCallbacks to sanitize outputs and BeforeModelCallbacks to reject unsafe input.

go-adk