🤖 feat: add mode="compact" for /compact operations (#200)

Enables users to customize AI compaction behavior via instruction files.

## What Changed

**Backend accepts mode as string** - No hardcoded mode restrictions,
allowing future flexibility without backend changes.

**Frontend sends mode="compact"** - When users run `/compact`, the
system now passes `mode: "compact"` to include mode-specific
instructions from AGENTS.md.

**Documentation** - Added example `Mode: Compact` section showing how
users can customize what information is preserved during automatic
conversation history compaction.

## Example Usage

Add to `~/.cmux/AGENTS.md` or `<workspace>/AGENTS.md`:

```markdown
## Mode: Compact

When compacting conversation history:

- Preserve key decisions and their rationale
- Keep code snippets that are still relevant  
- Maintain context about ongoing tasks
- Be extremely concise—prioritize information density
```

When the user runs `/compact`, the AI will receive these additional
instructions and follow them during summarization.

## Technical Notes

- Backend now accepts `mode?: string` for flexibility
- Frontend uses type-safe literal: `mode: "compact" as const`
- Available modes: exec, plan, compact (extensible without backend
changes)

_Generated with `cmux`_

Co-authored-by: Ammar Bandukwala <ammar@ammar.io>

Author: ammar-agent

docs/instruction-files.md

@@ -39,8 +39,25 @@ When planning:
 - Focus on goals, constraints, and trade-offs
 - Propose alternatives with pros/cons
 - Defer implementation detail unless asked
+
+## Mode: Compact
+
+When compacting conversation history:
+
+- Preserve key decisions and their rationale
+- Keep code snippets that are still relevant
+- Maintain context about ongoing tasks
+- Be extremely concise—prioritize information density
 ```
 
+### Available modes
+
+- **exec** - Default mode for normal operations
+- **plan** - Activated when the user toggles plan mode in the UI
+- **compact** - Automatically used during `/compact` operations to guide how the AI summarizes conversation history
+
+Customizing the `compact` mode is particularly useful for controlling what information is preserved during automatic history compaction.
+
 ## Practical layout
 
 ```

src/components/ChatInput.tsx

@@ -625,6 +625,7 @@ export const ChatInput: React.FC<ChatInputProps> = ({
               thinkingLevel: isAnthropic ? "off" : sendMessageOptions.thinkingLevel,
               toolPolicy: [{ regex_match: "compact_summary", action: "require" }],
               maxOutputTokens: parsed.maxOutputTokens, // Pass to model directly
+              mode: "compact" as const, // Allow users to customize compaction behavior via Mode: compact in AGENTS.md
             });
 
             if (!result.success) {

src/services/aiService.ts

@@ -390,7 +390,7 @@ export class AIService extends EventEmitter {
    * @param additionalSystemInstructions Optional additional system instructions to append
    * @param maxOutputTokens Optional maximum tokens for model output
    * @param cmuxProviderOptions Optional provider-specific options
-   * @param mode Optional UI permission mode ("exec" | "plan") - affects system message
+   * @param mode Optional mode name - affects system message via Mode: sections in AGENTS.md
    * @returns Promise that resolves when streaming completes or fails
    */
   async streamMessage(
@@ -403,7 +403,7 @@ export class AIService extends EventEmitter {
     additionalSystemInstructions?: string,
     maxOutputTokens?: number,
     cmuxProviderOptions?: CmuxProviderOptions,
-    mode?: "exec" | "plan"
+    mode?: string
   ): Promise<Result<void, SendMessageError>> {
     try {
       if (this.mockModeEnabled && this.mockScenarioPlayer) {

src/types/ipc.ts

@@ -140,7 +140,7 @@ export interface SendMessageOptions {
   additionalSystemInstructions?: string;
   maxOutputTokens?: number;
   providerOptions?: CmuxProviderOptions;
-  mode?: "exec" | "plan"; // UI permission mode
+  mode?: string; // Mode name - frontend narrows to specific values, backend accepts any string
 }
 
 // API method signatures (shared between main and preload)