add-backend

name	add-backend
description	Scaffold a new LLM backend for multi-bark-pack
user-invocable	true

Add Backend Skill

Scaffold a new LLM agent backend for multi-bark-pack.

Usage

/add-backend cursor
/add-backend codex
/add-backend <backend-name>

Steps to Add a New Backend

1. Create Backend Module

Create backends/<name>.js:

/**
 * <Name> CLI Backend
 */

const { execSync } = require('child_process');
const crypto = require('crypto');

const EXEC_OPTS = { encoding: 'utf8', timeout: 5000 };

module.exports = function create<Name>Backend(config = {}) {
    return {
        name: '<name>',
        displayName: '<Display Name>',

        async isInstalled() {
            try {
                execSync('which <cli-command>', EXEC_OPTS);
                return true;
            } catch {
                return false;
            }
        },

        async getVersion() {
            try {
                const output = execSync('<cli-command> --version', EXEC_OPTS);
                return output.trim();
            } catch {
                return null;
            }
        },

        models: ['model1', 'model2'],
        defaultModel: 'model1',

        validateModel(model) {
            return this.models.includes(model);
        },

        canResume: true,

        generateSessionId() {
            return crypto.randomUUID();
        },

        buildCommand(opts) {
            const { promptFile, sessionId, isResume, model, systemPromptFile, streamParserScript, agentId, tmpDir } = opts;

            // Build CLI-specific command
            let script = '#!/bin/bash\n';
            script += `# TODO: Implement <name> CLI invocation\n`;

            return { script, env: {} };
        },

        streamParserName: '<name>',

        extractSessionId(output) {
            return null;
        },

        capabilities: {
            streaming: true,
            sessionPersistence: true,
            workingDirectory: true,
            forceMode: true,
            systemPrompt: true,
            planning: true,
        },
    };
};

2. Create Stream Parser

Create stream-parsers/<name>.js:

/**
 * <Name> stream parser
 */

const TOOL_ICONS = {
    // Map tool names to icons
};

module.exports = {
    name: '<name>',
    toolIcons: TOOL_ICONS,

    parseLine(line) {
        // Parse CLI output format
        // Return: { type: 'text'|'tool'|'result', ... }
    },

    getToolIcon(name) {
        return TOOL_ICONS[name] || '🔧';
    },
};

3. Register Backend

In backends/index.js, add:

const create<Name>Backend = require('./<name>');

const backendFactories = {
    'claude-code': createClaudeCodeBackend,
    '<name>': create<Name>Backend,  // Add this
};

4. Register Parser

In stream-parsers/index.js, add:

const <name>Parser = require('./<name>');

const parsers = {
    claude: claudeParser,
    '<name>': <name>Parser,  // Add this
};

5. Update Config

In .env.example, add:

ENABLED_BACKENDS=claude-code,<name>

6. Test

node server.js
# In chat: /backends (should show new backend)
# Spawn a pup with new backend (once implemented)

Checklist

• Backend module in backends/<name>.js
• Stream parser in stream-parsers/<name>.js
• Registered in backends/index.js
• Registered in stream-parsers/index.js
• CLI availability check works
• Session management works
• Streaming output parsed correctly
• Updated docs in CLAUDE.md