This is the coding skill in the Google agents-cli series, providing a quick reference for ADK: agent types, tool definitions, orchestration patterns, callbacks, state management, and the graph-based Workflow API. For Python, start by reading the recipes index; ready-made implementations include sandboxed code execution, skills loadable from SKILL.md, cross-session memory, approvals before high-risk operations, tool guardrails, per-user credentials, and scheduled or event-driven triggers. For Go, refer to the examples in adk-go. The SKILL.md requires scaffolding the project before writing code, with deployment handled by the deploy skill in the same series.
Why it matters Before adding tools or writing callbacks in an ADK project, agents should check existing examples rather than inventing APIs from memory. Implementations for approval gates, cross-session memory, and more are available to copy.
skills.sh installs 572414 (+2199 / 24h)
Security Audit: Risk warnings present
- Installs or runs third-party packages (npx / pip, etc.), which involves executing commands
- Makes network requests or calls external APIs / remote MCP
- Audit scope: GitHub google/agents-cli/skills/google-agents-cli-adk-code/SKILL.md (only the SKILL.md text was scanned)
Rationale
- skills.sh installs 24 hours 2199 (indicates attention only); Apache-2.0 license
- The third-party package installation flagged in the audit is
uv tool install google-agents-cli, and the network activity is reading upstream Go examples, which aligns with the intended use
Usage
- Install the CLI using
uv tool install google-agents-cli, then instruct the agent in an ADK project to "add a tool" or "create a workflow with approvals"
Risks
- Requires enabling the workflow skill from the same series and scaffolding the project with agents-cli; deploying to Google Cloud requires a corresponding account