Cloudera Docs

Coding assistant tools reference

A comprehensive reference guide detailing the capabilities, parameters, and constraints of the tools available within the Agent Studio sandbox environment.

File Reader

The File Reader tool extracts contents from files on the local filesystem and formats them with prefixed line numbers, similar to the standard cat -n output format.
  • Parameters - Accepts absolute file paths along with optional offset and limit parameters to enable efficient pagination through large files.
  • Supported formats - It processes PDF files - up to 20 pages per request, requiring pagination for files exceeding 10 pages and Jupyter notebooks (.ipynb), returning all cells and their outputs in a unified response.
  • Scope - Operations are restricted to the sandbox's persistent /workspace directory.

File Writer

The File Writer tool is optimized for creating new files or performing complete overwrites of existing files within the sandbox filesystem.
  • Path automation - Accepts an absolute file path and automatically creates any missing intermediate parent directories.
  • Capacity limits - Enforces a strict threshold of approximately 80 lines of content per call to prevent output token saturation and silent text truncation.
  • Data persistence - Files must be designated within the /workspace/<filename> path structure to ensure availability across different steps in an active session.

Find and Replace

The Find and Replace tool runs precise, localized string modifications inside existing source files. It serves as the primary mechanism for targeted edits, minimizing token usage compared to complete file rewrites.

  • Tool run - Given a file path, an old_string, and a new_string, the tool atomically modifies the target text. Enabling the replace_all flag enables global substitution of a specific variable or identifier throughout the entire file.
  • Strict matching - The tool requires an exact match of the target string, including whitespace and specific indentation levels and displays an error if the pattern cannot be located.

Bash Executor

The Bash Executor tool runs arbitrary shell commands within the Agent Studio sandbox environment, capturing and returning both standard output (stdout) and standard error (stderr).
  • Capabilities - Provides functionality for running scripts and test suites, installing software packages, and verifying environment runtime configurations.
  • Environment constraints -
    • The sandboxed space is Python-only.
    • Runtimes for languages such as Node.js or Ruby are unsupported.
    • External network access is permitted, but file writes outside of /workspace is prohibited.
  • Package management - Operates using an optimized uv framework. A virtual environment is initiated at /workspace/.venv on first use, packages are provisioned using the uv pip install command, and scripts run directly through /workspace/.venv/bin/python.
  • Data visualization - Data plotting is supported through matplotlib. Plots must be exported directly to the file system using plt.savefig(), as there is no graphical user interface display inside the sandbox.

Directory Explorer

The Directory Explorer tool lists the structures of specified directories or locates files matching explicit glob patterns, such as **/*.py. Returns file names, directory structures, and file-type indicators, but does not read or expose the underlying file content.

It is for standard codebase mapping, directory structural listing, and verifying that workspace commands or scripts have successfully written output artifacts.

Code Search

The Code Search tool uses standard regular expressions, and runs comprehensive, high-performance textual searches across the codebase.
  • Functionality - Locates variable usages, function definitions, imports, or complex structural matches, returning the corresponding file paths, line numbers, and matching text lines.
  • Scoping - Accepts an optional path argument to focus the search boundaries to specific subdirectories or file globs, avoiding irrelevant results. Supports searching within the active session's generated /workspace directory.

Lint Checker

The Lint Checker tool evaluates code health and syntax correctness by running static analysis engines across selected files or directories.

  • Language Support - Features native validation for Python, through ruff or flake8) and JavaScript or TypeScript, through eslint.
  • Diagnostics - Detects and returns syntax errors, style formatting problems, and warnings alongside exact file coordinates, such as paths and line numbers.

Web Fetch

The Web Fetch tool downloads web documents from publicly accessible URLs and cleanly converts them into readable, structured markdown text.

  • Optimized processing - Seamlessly handles URL redirects and automatically upgrades HTTP requests to HTTPS, while filtering out website headers, footers, sidebars, and navigation bars to provide clean prose and formatted code blocks.
  • Access restrictions - Designed for open resources such as library documentation, public GitHub files, or package repositories. It does not support private or authenticated URLs, such as internal Confluence spaces, Jira boards, or private code repositories.

Web Search

The Web Search tool runs automated searches to retrieve organized and prioritized data, which features page titles, target URLs, and context-rich summaries, using DuckDuckGo.
  • Use case - Retrieves real-time technical information outside the core system training scope, such as contemporary library updates, new package releases, or community troubleshooting threads.
  • Filters - Supports strict domain control using allowlist (allowed_domains) and blocked lists (blocked_domains) to focus search relevancy.
  • Security - All tool actions take place inside an isolated, sandboxed space. Persistent files, output logs, and virtual environments remain stable throughout the active session within the designated /workspace directory.

Email Tool

The Email Tool sends emails by connecting to an SMTP server. This toll requires the following configurations:
  • SMTP Server - The address of the SMTP server.
  • SMTP Port – The port number for the SMTP connection.
  • SMTP Password – The password required for authenticating with the SMTP server.

Cloudera Data Flow HuggingFace Readyflow Deploy Tool

The Cloudera Data Flow HuggingFace Readyflow Deploy Tool imports a huggingface dataset into a path in the Cloudera Data Lake.

This tool requires the following configurations:

  • Cloudera workload username – The username for accessing Cloudera workloads.
  • Cloudera API Access Key – Credential for authentication with the Cloudera API.
  • Cloudera API Private Key – Secret key for secure API access.
  • CRN of a running Cloudera Data Flow service – The unique identifier of an active Cloudera Data Flow service instance.

Cloudera Data Visualization Dataset Creation Tool

The Cloudera Data Visualization Dataset Creation Tool generates a Cloudera Data Visualization dataset directly from an existing SQL table.

This tool requires the following configurations:

  • Dataset connection ID – The identifier for the data connection in Cloudera Data Visualization.
  • Cloudera Data Visualization Base URL – The URL of your Cloudera Data Visualization instance.
  • Cloudera AI APIv2 application key – Credential for authentication with Cloudera AI APIs.

Cloudera Data Visualization Smart Visual Tool

The Cloudera Data Visualization Smart Visual Tool generates visualization within Cloudera Data Visualization Smart Visual Tool, using a specified dataset. This tool requires the following configurations:
  • Cloudera Data Visualization Base URL – The URL of your Cloudera Data Visualization instance.
  • Cloudera AI APIv2 Application Key – Credential for authentication with Cloudera AI APIs.

Cloudera Data Warehouse Hive Table from S3 Parquet Tool

The Cloudera Data Warehouse Hive Table from S3 Parquet Tool creates a table in Cloudera Data Warehouse by loading an existing parquet file from S3. This tool validates the successful creation of the table by returning sample data from the newly accessible table.

This tool requires the following configurations:

  • Workload user – The username required for Cloudera Data Warehouse workload access.
  • Workload password – The corresponding password for authentication.
  • Hive Cloudera AI data connection name – The name of the data connection linking Hive to the Parquet file located in S3.