valkryhx/google_adk_agent / skills/programmatic-tool-calling
programmatic-tool-calling
Verified编程化工具调用指南。允许 Agent 编写并执行 Python 代码,从而能够在一个沙箱环境中动态调用其他工具,实现循环、批处理和复杂逻辑。
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 5.2 KB |
| SKILL.md.metadata.json | config | 158 B |
| SKILL.md.resolved | other | 2.6 KB |
| assets/example_asset.txt | markdown | 865 B |
| references/api_reference.md | markdown | 976 B |
| scripts/example.py | script | 609 B |
| tools.py | script | 3.9 KB |
SKILL.md.resolved
View on GitHub---
name: programmatic-tool-calling
description: Guide and reference for using Programmatic Tool Calling with Claude, enabling the model to write Python code that invokes other tools.
---
# Programmatic Tool Calling
This skill provides documentation and patterns for "Programmatic Tool Calling", a feature where Claude writes Python code to invoke tools instead of the user (or orchestration layer) calling them directly.
## Overview
Programmatic tool calling allows Claude to:
1. Write Python code (running in a sandboxed environment).
2. Call other tools *from within* that Python code.
3. Process the results programmatically (loops, conditionals, filtering) before returning the final answer.
## Key Concepts
### 1. `allowed_callers`
When defining a tool, you can specify who is allowed to call it via the `allowed_callers` field in the tool definition.
- `["direct"]`: Only Claude can call this tool directly (standard tool use).
- `["code_execution_20250825"]`: Only callable from within the code execution sandbox.
- `["direct", "code_execution_20250825"]`: Callable by both.
**Best Practice**: Choose one or the other to provide clear guidance to the model.
### 2. The `caller` field
When a tool is called, the API response includes a `caller` field:
- `{"type": "direct"}`: Traditional invocation.
- `{"type": "code_execution_20250825", "tool_id": "srvtoolu_..."}`: Programmatic invocation from a specific code execution block.
## Workflow
1. **Orchestrator** sends a message to Claude with tools defined (some with `allowed_callers=["code_execution_20250825"]`).
2. **Claude** decides to use a tool but realizes it must be called via code.
3. **Claude** generates a `tool_use` block for `code_execution` (e.g., writing a Python script).
4. **Orchestrator** executes the Python code.
5. **Python Code** hits a function call (e.g., `await query_database(...)`).
6. **Orchestrator** pauses code execution, sends a `tool_use` for `query_database` to the API (or handles it locally if possible).
7. **System** executes `query_database` and returns the result to the *Python environment*.
8. **Python Code** resumes with the result.
9. **Claude** receives the final output from the code execution.
## Example Use Cases
- **Batch Processing**: Query a database for 100 records and filter them in Python before showing them to the user.
- **Multi-step Logic**: "If API A returns X, call API B; else call API C."
- **Data Aggregation**: Call a search tool 5 times in parallel (using async) and summarize the results.
## Reference
See `https://platform.claude.com/docs/en/agents-and-tools/tool-use/programmatic-tool-calling` for full documentation.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| reverse-engineer-anythingmorluto/reaReverse engineer native, managed, Electron/JavaScript, packaged, firmware, and browser targets with REA, and build agent environments (AgentEnv bundles) of web applications from a live instance or a capture, and of Linux desktop apps with a task from a request. Use shipped-artifact or requested runtime evidence to explain features, compare versions, decompile code, or guide a reconstruction. Skip REA for ordinary source-repository architecture analysis. | Stars: 471k | Updated | Claude CodeCodex | Verified | Install |
| diagnosing-superpowersobra/superpowersUse when a superpowers session went wrong and your human partner wants to know why — repeated work, ignored plans, stumbles, poor results, a skill that didn't fire, "it took too long", "why is it so expensive", "what is it doing" — or wants to build a bug report for the superpowers maintainers, for the current session or a past one identified by id or path, on any harness. | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| dispatching-parallel-agentsobra/superpowersUse when facing 2+ independent tasks that can be worked on without shared state or sequential dependencies | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| executing-plansobra/superpowersUse when executing an implementation plan in the current session as the implementer yourself — your human partner chose inline execution, or no subagent tool is available | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |