Create and manage @tool decorated functions for ReinforceNow agents.
npx skills add https://github.com/ReinforceNow/reinforcenow-cli --skill rnow-tools
Please help me install this Agent Skill. Skill: rnow-tools Source: https://github.com/ReinforceNow/reinforcenow-cli/tree/main/project8/.claude/skills/rnow-tools Command: npx skills add https://github.com/ReinforceNow/reinforcenow-cli --skill rnow-tools
This skill helps you create and manage reusable tool functions for ReinforceNow agents, streamlining how the model calls external logic during training and inference.
Create a tools.py file in your project, decorate functions with @tool, supply a descriptive docstring, and return JSON-serializable data. For sandboxed tools, decorate with @tool(sandbox=True, timeout=120) and include docker image:tag in your train.jsonl entries.
High-intent search queries and answers about installing and using this skill.
To create reusable agent training tools, define functions in a tools.py module, decorate them with @tool, add explicit type hints and docstrings, and ensure all return values are JSON-serializable for reliable model execution.
Enable sandbox mode for stateful agent tools when executing logic requires isolation. Decorate functions with @tool(sandbox=True, timeout=120) and specify a docker image:tag in your train.jsonl entries to maintain deterministic stateful environments.
Decorating functions with @tool in a tools.py module requires explicit parameter type hints, descriptive docstrings for model comprehension, and strictly JSON-serializable return values to ensure proper agent function calling during training and inference.
Update train.jsonl entries for sandboxed docker tools by appending the required docker image:tag configuration to each training data record, ensuring the agent training environment aligns with the isolated execution context defined by the @tool(sandbox=True) decorator.
Yes, you can build stateless API wrappers and calculators without sandbox isolation by applying the standard @tool decorator without the sandbox parameter, provided the functions return JSON-serializable data and include proper type annotations and docstrings.
@tool decorated functions fail during agent training if they lack explicit type hints, miss descriptive docstrings, return non-JSON-serializable data, or have missing docker image:tag references in train.jsonl when sandbox mode is enabled.
gujiachun-rainbowCreate or upgrade callable Python tools using the @tool decorator.
ReinforceNowFormat and validate train.jsonl entries for ReinnowNow RL and SFT datasets.
databricks-solutionsOrchestrate tool-calling agent loops with Databricks-hosted Foundation Models.
markus41Document the Claude Code configuration schema for permissions, hooks, and environment variables.
