pi-compact-display

An extension for the Pi Assistant that allows you to finely customize the terminal (TUI) output for various tools (e.g., bash, read, ls, edit) on a per-tool basis.

It helps keep your context and screen clean by limiting overly long tool outputs or hiding unnecessary tool calls, summarizing them instead.

Features

• Per-Tool Configuration: Set different display modes for each tool.
• Summary Display: Hide execution details and output only a summary like ⚡ read(1) ls(2) at the top of the assistant's response.
• Line Limit & Padding Removal: Limit the number of output lines and automatically strip empty padding lines.
• Grouped Tool Results: With grouping enabled, all lines-mode calls in one user turn are combined into a single aggregation line (e.g. ⚡ bash ×3).
• Suppress Thinking Indicator: With hideThinking enabled, the "Thinking..." indicator (and full thinking blocks) shown for thinking models are hidden from the assistant messages.
• Support for External Tools: Override the display settings not only for Pi's built-in tools but also for any custom tools added by other extensions.
• Expandable Details: Fully supports Pi's native expand feature (e.g., pressing Ctrl+O in the TUI), overriding the limits to show the full output.

---

Installation & Build

From npm (Recommended)

You can install the extension directly from npm:

pi install npm:pi-compact-display

From Source (Local Development)

# Clone or navigate to the repository
cd pi-compact-display

# Install dependencies
npm install

# Build the TypeScript code
npm run build

# Install the extension using the Pi CLI
pi install .

---

Configuration

Create a configuration file named config.json at the following location:

File Path: ~/.pi/agent/extensions/pi-compact-display/config.json

Configuration Structure

The configuration is in JSON format, where the key is the "tool name" and the value is its "display settings (options)".

{
  "tool_name": {
    "mode": "display_mode",
    "outputLines": number,
    "noPadding": true or false
  }
}

Available Properties

Property	Type	Description
mode	string	The display mode for the tool. Must be "count_only", "lines", or "default". (Required)
outputLines	number	The maximum number of lines to display. Only effective when mode is "lines". (Optional)
noPadding	boolean	Whether to remove leading, trailing, and consecutive empty lines from the output. Set to true to omit empty lines. Only effective when mode is "lines". When grouped, the overall padding of the group card follows the noPadding setting of the leader (the first tool in the group). (Optional)
grouping	boolean	Whether to include this tool in the group card when the global grouping setting is enabled. Set to false to exclude it from grouping and display it as a separate card. Default is true. (Optional)

Display Modes (mode)

1. "count_only"

   • Behavior: The tool execution and its output are completely hidden from the screen. Instead, the number of executions is summarized at the top of the assistant's message (e.g., ⚡ read(1)).
   • Use Case: Best for tools like read, ls, find, or grep, where the LLM gathers information autonomously and humans do not need to read the raw output.

2. "lines"

   • Behavior: Displays the tool execution command and a truncated portion of its result. It limits the output to the specified outputLines.
   • Use Case: Ideal for tools like bash, edit, or write, where humans want to quickly check the executed command and whether it resulted in an error.

3. "default"

   • Behavior: Bypasses this extension's rendering overrides and falls back to Pi's default rendering logic (or logic defined by other extensions).
   • Note: Any tool not specified in the configuration file, or if config.json is missing entirely, will automatically default to default.

---

Example Configuration

{
  "grouping": true,
  "hideThinking": true,
  "user": {
    "noPadding": true
  },
  "skill": {
    "noPadding": true
  },
  "read": { "mode": "count_only" },
  "ls": { "mode": "count_only" },
  "find": { "mode": "count_only" },
  "grep": { "mode": "count_only" },
  
  "bash": { 
    "mode": "lines", 
    "outputLines": 3, 
    "noPadding": true 
  },
  "write": { 
    "mode": "lines", 
    "outputLines": 1, 
    "noPadding": true 
  },
  
  "mcp:tavily_tavily_search": { "mode": "count_only" },
  "mcp__tavily:tavily_search": { "mode": "lines", "outputLines": 5 },
  "mcp": { "mode": "lines", "outputLines": 10 },
  "default": { "mode": "count_only" },

  "my_custom_tool": {
    "mode": "count_only"
  }
}

Note: You can specify any tool name, including custom tools added by other extensions like my_custom_tool.

Sub-tool and Action-Specific Configuration (Gateway Tools like MCP)

For "gateway tools" that call other tools or actions internally (such as the mcp tool), you can configure them specifically based on their arguments.

The configuration priority is evaluated in a cascading fallback: Specific Sub-tool Configuration > General Tool Configuration > default Key Configuration > System Default. This mechanism applies to all tools, not just MCP.

How the sub-tool name is derived (the same rule for every tool):

• If the argument tool is a string → <tool_name>:<value> (e.g. mcp:tavily_search when called with { "tool": "tavily_search" })
• If the argument action is a string → <tool_name>:<value> (e.g. mcp:install)
• No tool-specific argument names are evaluated. mcp has no special case: a call that only carries its own argument names (e.g. { "server": "..." }, { "connect": "..." }) falls back to the general mcp key.

Server proxy names (namespace proxies): Depending on how your MCP servers are exposed, the tool name itself may already be namespaced. Those names are also usable as configuration keys:

• mcp__<server_name> (e.g. mcp__tavily, or mcp__playwright_mcp_chrome for a server named playwright-mcp-chrome)
• mcp__<server_name>:<tool_name> (e.g. mcp__tavily:tavily_search)

The count_only summary line displays exactly this name, so you can copy it straight into the configuration as a key. In lines mode, the call line displays the invocation as <effective_name> { <arguments> } (e.g. mcp__tavily:tavily_search { query: "Node.js 22" }), matching the effective name used for configuration.

Note: The same priority (specific setting > "default" setting) applies to built-in tools like read and bash as well.

User Prompt Padding Configuration

If you want to remove the vertical empty lines (vertical padding) that appear above and below the prompt messages you enter, you can configure them specifically using the "user" key. This configuration is not affected by other configurations like "default" and is only applied if explicitly defined.

Configuration Example:

{
  "user": {
    "noPadding": true
  }
}

• noPadding (boolean): Set to true to remove the empty lines (padding) inserted above and below your prompt input, making the UI extremely compact.

Skill Invocation Padding Configuration

If you want to remove the vertical empty lines (vertical padding) that appear above and below the [skill] display (rendered when you invoke a skill via /skill:... or a <skill> block), you can configure them specifically using the "skill" key. Like "user", this configuration is not affected by other configurations such as "default" and is only applied if explicitly defined.

Configuration Example:

{
  "skill": {
    "noPadding": true
  }
}

• noPadding (boolean): Set to true to remove the empty lines (padding) inserted above and below the [skill] block, making the UI extremely compact.

Global Setting: Grouping Tool Results (grouping)

lines-mode tools normally render one tool card per call. When this is enabled, all lines-mode calls within one user turn are combined into a single card that shows only the aggregation header, greatly reducing the vertical space used on screen. Even if the assistant splits its work into multiple thinking passes (multiple assistant messages), everything in the same user turn stays in one card.

Configuration Example:

{
  "grouping": true,
  "bash": { "mode": "lines", "outputLines": 3, "grouping": false },
  "write": { "mode": "lines", "outputLines": 1 }
}

• grouping (boolean): Set to true to combine all lines-mode calls in one user turn (regardless of tool name) into a single card. The card shows only the count header such as ⚡ bash ×3 edit ×1 (even a single call is shown as ⚡ bash ×1). Like "user", this setting is not affected by other configurations such as "default" and is only applied if explicitly defined. You can exclude specific tools from the group by setting "grouping": false in their individual tool configuration.
• Command lines and outputs are hidden by default; press Ctrl+O (expand) to show every call's command line and full, untruncated output for all tools in the group.
• Calls in different turns (separated by a user message) are rendered as separate cards.

Global Setting: Suppress Thinking Indicator (hideThinking)

Thinking-capable models may emit thinking content blocks, which Pi renders as a "Thinking..." label (or as the full thinking text when thinking display is toggled on). When this setting is enabled, those thinking blocks are filtered out of the assistant message entirely, so no "Thinking..." indicator (and no thinking content) appears on screen.

Configuration Example:

{
  "hideThinking": true
}

• hideThinking (boolean): Set to true to suppress the "Thinking..." indicator and all thinking content in assistant messages. Like "user" and "grouping", this setting is not affected by other configurations such as "default" and is only applied if explicitly defined.

---

Usage & Shortcuts

After setting up the configuration, simply start the Pi assistant.

• Context Cleanup: Tools set to count_only will display summaries like ⚡ read(1) on the screen. However, this summary text is automatically stripped from the conversation history (context) sent to the LLM behind the scenes, so it won't pollute the LLM's context window.
• Full Output (Expanded): If the output is truncated in lines mode, you can use Pi's native expand shortcut (typically Ctrl+O) to bypass the line limits and view the full output.