DeepSeek-Reasonix Output Formats: Built-In Styles and Custom Configuration

DeepSeek-Reasonix supports three built-in output formats—explanatory, learning, and concise—plus unlimited custom styles defined via Markdown files in your project's .reasonix/output-styles/ directory.

DeepSeek-Reasonix is an open-source coding assistant that adapts its communication style to match your workflow needs. The tool provides flexible output formats that control how analysis is presented, from verbose explanations to terse code-only responses. You can switch between these formats using the /output-style command or define your own custom personas to optimize the analysis experience.

Built-In Output Styles

The internal/outputstyle/outputstyle.go file defines three core personas that ship with DeepSeek-Reasonix. These styles are automatically available in every session and append specific instructions to the system prompt when activated.

Explanatory Style

The explanatory format generates detailed analysis with insight blocks. When active, the system appends instructions that trigger short "## Insight" notes after substantive code changes, explaining non-obvious implementation choices as they occur. This style keeps the coding flow active (keep-coding = true) while adding educational context.

Learning Style

The learning format transforms the assistant into a collaborative tutor. It inserts TODO(human) stubs at decision points that require user input, pausing execution to let you complete critical sections. Like the explanatory style, it maintains keep-coding = true, but structures the output to guide hands-on learning rather than providing complete solutions.

Concise Style

The concise format minimizes prose and maximizes code density. It strips preambles and post-ambles, delivering only essential bullets and code blocks. This format is ideal for experienced developers who want rapid answers without explanatory fluff, keeping the coding instructions active while enforcing brevity.

Custom Output Styles

You can extend DeepSeek-Reasonix beyond its built-in personas by creating custom output styles. Place any Markdown file with a .md extension in the ./.reasonix/output-styles/ directory (or the home directory equivalent) to register a new format.

Custom styles support YAML front-matter to configure behavior:

• name: The style identifier (case-insensitive match against built-ins)
• description: Shown when listing available styles
• keep-coding-instructions: Boolean flag indicating whether to replace the prompt (false) or append to it (true)

If a custom style name matches a built-in style name, the custom version takes precedence. When no style is selected (empty string or default), the system prompt remains unchanged.

How to Switch Between Output Formats

Use the /output-style command in the interactive terminal UI to manage output formats. The implementation in internal/cli/chat_tui.go (lines 3338-3344) handles this command by displaying the current active style and allowing you to switch contexts.

To list all available styles:

/output-style

This displays built-in and custom styles with their descriptions:

* concise (builtin) — Terse replies: minimal prose, code and bullets only
  explanatory (builtin) — Explain non‑obvious implementation choices as you go
  learning (builtin) — Collaborate and leave TODO(human) stubs for the user to complete
  my-custom (custom) — My personal verbose persona

To activate a specific style:

/output-style learning

The session immediately adopts the learning persona, inserting TODO(human) markers wherever choices are deferred to the user.

Creating a Custom Output Style

To add a custom persona, create a Markdown file with front-matter in your project's output-styles directory:

---
name: verbose
description: Very detailed explanations, with intermediate reasoning
keep-coding-instructions: true
---
Communication style — Verbose: always include the full chain‑of‑thought, even for trivial steps. Use paragraphs and bullet lists liberally.

After saving the file, run /output-style again. The verbose style appears under "custom" and can be selected immediately.

Implementation in the Source Code

The output format system is implemented across several key files in the esengine/DeepSeek-Reasonix repository.

The internal/outputstyle/outputstyle.go file contains the core logic:

• Built-in definitions: Lines 33-60 hard-code the three default styles
• Discovery functions: The List function (lines 81-106) and Resolve function (lines 112-125) scan directories returned by outputstyle.Dirs() to find custom .md files, parse their front-matter, and merge them with built-ins

The runtime logic prioritizes user-defined styles over built-ins when names conflict, handling case-insensitive matching automatically.

Summary

• DeepSeek-Reasonix output formats include three built-in styles: explanatory (insight-rich), learning (collaborative with TODOs), and concise (minimal prose).
• Custom styles are created by placing Markdown files with front-matter in ./.reasonix/output-styles/<name>.md.
• The keep-coding-instructions flag determines whether custom styles append to or replace the default system prompt.
• Use the /output-style command in the interactive UI to list and switch between available formats.
• Core implementation resides in internal/outputstyle/outputstyle.go, with UI handling in internal/cli/chat_tui.go.

Frequently Asked Questions

What are the three built-in DeepSeek-Reasonix output formats?

DeepSeek-Reasonix ships with explanatory, learning, and concise output formats. The explanatory style adds insight blocks to explain implementation choices, the learning style inserts TODO(human) stubs for collaborative coding, and the concise style delivers minimal prose with code-first responses. All three are defined in internal/outputstyle/outputstyle.go and set keep-coding = true.

How do I create a custom output style in DeepSeek-Reasonix?

Create a Markdown file in ./.reasonix/output-styles/<name>.md (or your home directory equivalent) with YAML front-matter specifying name, description, and keep-coding-instructions. The file should contain the persona instructions in Markdown format. The List and Resolve functions in internal/outputstyle/outputstyle.go automatically discover and load these files at runtime.

Can custom output styles override the built-in formats?

Yes. If a custom style file name matches a built-in style name (case-insensitive), the custom version takes precedence. You can use this to customize the behavior of the explanatory, learning, or concise personas without renaming them.

Where is the output format logic implemented in the DeepSeek-Reasonix codebase?

The core logic lives in internal/outputstyle/outputstyle.go, specifically lines 33-60 for built-in definitions and lines 81-125 for the List and Resolve discovery functions. The interactive command handling is implemented in internal/cli/chat_tui.go around lines 3338-3344, which processes the /output-style command in the terminal UI.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}