Example Agents

"""
TAB Agent — Python Reference Implementation
=============================================
Implement the 'respond' function below. TAB will call it with each
benchmark prompt and use your return value as the agent's answer.
"""
import json
import re


def respond(prompt: str) -> str:
    """
    Main entry point called by TAB for every benchmark prompt.

    Args:
        prompt: The benchmark question or task as a plain string.

    Returns:
        Your agent's answer as a plain string.
    """
    prompt_lower = prompt.lower().strip()

    # ---- Example: simple keyword routing ----
    if "summarize" in prompt_lower or "summary" in prompt_lower:
        return summarize(prompt)

    if "extract" in prompt_lower:
        return extract_entities(prompt)

    # Default: echo-style fallback (replace with your real logic)
    return f"Received prompt ({len(prompt)} chars). Implement your logic here."


# --------------- helper functions ---------------

def summarize(text: str) -> str:
    """Produce a short summary of the input text."""
    sentences = re.split(r'(?<=[.!?])\s+', text)
    # Return first two sentences as a naive summary
    return " ".join(sentences[:2]) if sentences else text[:200]


def extract_entities(text: str) -> str:
    """Extract simple entities (numbers, emails) from text."""
    numbers = re.findall(r'\b\d+\.?\d*\b', text)
    emails = re.findall(r'[\w.+-]+@[\w-]+\.[\w.]+', text)
    result = {"numbers": numbers[:10], "emails": emails[:5]}
    return json.dumps(result, indent=2)

/**
 * TAB Agent — JavaScript / TypeScript Reference Implementation
 * =============================================================
 * Export a 'handler' function. TAB calls it with each benchmark
 * prompt and uses the return value as your agent's answer.
 */

/**
 * @param {string} prompt - The benchmark question or task.
 * @returns {string} Your agent's answer.
 */
function handler(prompt) {
    const lower = prompt.toLowerCase().trim();

    // ---- Example: keyword routing ----
    if (lower.includes("translate")) {
        return translate(prompt);
    }

    if (lower.includes("classify") || lower.includes("categorize")) {
        return classify(prompt);
    }

    // Default fallback (replace with your real logic)
    return `Received prompt (${prompt.length} chars). Implement your logic here.`;
}

// --------------- helper functions ---------------

function translate(text) {
    // Placeholder: real agent would call a translation API
    return "Translation placeholder — integrate your translation service here.";
}

function classify(text) {
    const categories = ["Technical", "Business", "Creative", "Support", "General"];
    // Naive keyword-based classification
    if (/code|api|bug|deploy/i.test(text)) return "Technical";
    if (/revenue|sales|market/i.test(text)) return "Business";
    if (/write|story|design/i.test(text)) return "Creative";
    if (/help|issue|problem/i.test(text)) return "Support";
    return "General";
}

module.exports = { handler };

my-agent/
├── main.py          # or main.js / main.ts — entry point
├── requirements.txt # Python deps (pip install) — optional
├── package.json     # Node deps (npm install)  — optional
└── utils/
    └── helpers.py   # Additional modules