Table of Contents

AI Agent Setup

DTK integrates with AI coding agents to automatically reduce token usage from dotnet commands.

Important

Python 3 requirement: The Claude Code and Gemini CLI integrations install Python-based hooks that run at command interception time. Make sure python3 is available on your PATH before using dtk integrate claude or dtk integrate gemini. Other providers (Copilot, Cursor, Windsurf, Aider, JetBrains) do not require Python.

Claude Code

A pre-built hook automatically rewrites dotnet build|test|restore|clean commands to use dtk.

Installation

From your project root, run:

dtk integrate claude

This creates three files:

  • .claude/skills/dotnet-token-killer/SKILL.md — instructs Claude Code to prefer dtk
  • .claude/hooks/dotnet-to-dtk.py — the Python rewrite hook
  • .claude/settings.json — registers the hook under PreToolUse (merges with any existing settings)

Re-running the command is safe: existing files are skipped. Use --force to overwrite:

dtk integrate claude --force

To target a directory other than the current one:

dtk integrate claude --dir /path/to/project

How It Works

With the hook in place, any time Claude Code runs dotnet build, dotnet test, dotnet restore, or dotnet clean, the command is silently rewritten to dtk dotnet ... before execution. The agent receives the filtered output without any extra configuration.

Manual Installation

If you prefer not to use dtk integrate, it requires curl and python3. From your project root:

mkdir -p .claude/hooks
curl -sSL https://raw.githubusercontent.com/HandyS11/DotnetTokenKiller/develop/.claude/hooks/dotnet-to-dtk.py \
  -o .claude/hooks/dotnet-to-dtk.py

Then add the following to .claude/settings.json:

{
  "hooks": {
    "PreToolUse": [
      {
        "matcher": "Bash",
        "hooks": [
          {
            "type": "command",
            "command": "python3 .claude/hooks/dotnet-to-dtk.py"
          }
        ]
      }
    ]
  }
}

GitHub Copilot (VS Code)

Installation

From your project root, run:

dtk integrate copilot

This appends a dtk instructions section to .github/copilot-instructions.md, creating the file if it does not exist. The section is wrapped in <!-- dtk --> / <!-- /dtk --> markers so re-running the command is safe. Use --force to refresh the section:

dtk integrate copilot --force

Manual Installation

Add to your .github/copilot-instructions.md:

Use `dtk` instead of raw `dotnet` for build, test, restore, and clean to reduce token usage.

```sh
dtk dotnet build
dtk dotnet test
dtk dotnet restore
dtk dotnet clean

Gemini CLI

A pre-built hook automatically rewrites dotnet build|test|restore|clean commands to use dtk.

Installation

From your project root, run:

dtk integrate gemini

This creates three files:

  • GEMINI.md — appends a dtk instructions section (creating the file if it does not exist)
  • .gemini/hooks/dotnet-to-dtk.py — the Python rewrite hook
  • .gemini/settings.json — registers the hook under BeforeTool (merges with any existing settings)

Re-running the command is safe: existing files are skipped. Use --force to overwrite:

dtk integrate gemini --force

To target a directory other than the current one:

dtk integrate gemini --dir /path/to/project

How It Works

With the hook in place, any time Gemini CLI runs dotnet build, dotnet test, dotnet restore, or dotnet clean, the command is silently rewritten to dtk dotnet ... before execution. The agent receives the filtered output without any extra configuration.

Manual Installation

If you prefer not to use dtk integrate, it requires curl and python3. From your project root:

mkdir -p .gemini/hooks
curl -sSL https://raw.githubusercontent.com/HandyS11/DotnetTokenKiller/develop/.gemini/hooks/dotnet-to-dtk.py \
  -o .gemini/hooks/dotnet-to-dtk.py

Then add the following to .gemini/settings.json:

{
  "hooks": {
    "BeforeTool": [
      {
        "matcher": "run_shell_command",
        "hooks": [
          {
            "type": "command",
            "command": "python3 .gemini/hooks/dotnet-to-dtk.py"
          }
        ]
      }
    ]
  }
}

And append the following to your GEMINI.md:

## DotnetTokenKiller (dtk)

Use `dtk` instead of raw `dotnet` for build, test, restore, and clean commands.
`dtk` filters output to actionable signal only, reducing noise by 50-97%.

Cursor

Installation

dtk integrate cursor

This creates .cursor/rules/dtk.mdc — a Cursor rule file with alwaysApply: false that instructs the agent to use dtk for dotnet commands.

Use --force to overwrite an existing file. Use --dir to target a specific project directory.

How It Works

Cursor loads .mdc rule files from .cursor/rules/ and applies them based on their alwaysApply setting. The generated rule tells the agent to prefer dtk dotnet build|test|restore|clean|format over raw dotnet commands. No hook or Python dependency is needed — it's a plain text instruction file.

Manual Installation

Create .cursor/rules/dtk.mdc:

---
alwaysApply: false
---

Use `dtk` instead of raw `dotnet` for build, test, restore, clean, and format commands.
`dtk` filters output to actionable signal only, reducing noise by 50–97%.

Windsurf

Installation

dtk integrate windsurf

This creates .windsurf/rules/dtk.md — a Windsurf rule file that instructs the agent to prefer dtk over raw dotnet commands.

Use --force to overwrite. Use --dir to target a specific project directory.

How It Works

Windsurf loads rule files from .windsurf/rules/ and applies them as system-level instructions. The generated file tells the agent to use dtk dotnet build|test|restore|clean|format to reduce token usage. No hook or Python dependency is needed.

Manual Installation

Create .windsurf/rules/dtk.md:

Use `dtk` instead of raw `dotnet` for build, test, restore, clean, and format commands.
`dtk` filters output to actionable signal only, reducing noise by 50–97%.

Aider

Installation

dtk integrate aider

This creates two files:

  • .aider-dtk-instructions.md — standalone instructions file referenced by Aider
  • .aider.conf.yml — a # dtk / # /dtk section is appended (creating the file if needed)

Re-running is safe; use --force to refresh the section.

Manual Installation

Add to your .aider.conf.yml:

# dtk
read:
  - .aider-dtk-instructions.md
# /dtk

And create .aider-dtk-instructions.md:

Use `dtk` instead of raw `dotnet` for build, test, restore, clean, and format commands.
`dtk` filters output to actionable signal only, reducing noise by 50–97%.

How It Works

Aider reads configuration from .aider.conf.yml, which can reference additional instruction files via the read: key. The integration adds a reference to .aider-dtk-instructions.md, which tells Aider to prefer dtk over raw dotnet commands. No hook or Python dependency is needed beyond Aider's own Python runtime.

JetBrains AI

Installation

dtk integrate jetbrains

This appends a <!-- dtk --> / <!-- /dtk --> instructions section to .junie/guidelines.md, creating the file if it does not exist. Re-running is safe; use --force to refresh the section.

How It Works

JetBrains AI (including Junie) reads project guidelines from .junie/guidelines.md. The integrated section instructs the agent to use dtk for all supported dotnet commands. The <!-- dtk --> markers allow safe re-generation without affecting other content in the guidelines file.

Manual Installation

Add to your .junie/guidelines.md:

<!-- dtk -->
## DotnetTokenKiller (dtk)

Use `dtk` instead of raw `dotnet` for build, test, restore, clean, and format commands.
`dtk` filters output to actionable signal only, reducing noise by 50–97%.

```sh
dtk dotnet build
dtk dotnet test
dtk dotnet restore
dtk dotnet clean
dtk dotnet format
```
<!-- /dtk -->

Other Agents

For any AI agent that runs terminal commands, the general approach is:

  1. Install DTK globally: dotnet tool install -g DotnetTokenKiller
  2. Configure the agent to prefix dotnet build|test|restore|clean with dtk
  3. The agent receives compact, filtered output — reducing token usage by 50–98%
Edit this page