Data Analysis Agent with OpenClaw

# agents/analyst-agent.yaml
name: analyst-agent
model: gpt-4o
temperature: 0.1
tools:
  - csv_reader
  - sql_tool
  - python_executor   # sandboxed Python for pandas/numpy
  - chart_generator   # produces chart JSON for Chart.js
system: |
  You are a data analyst agent. Think step by step.
  - Load the data, inspect shape and dtypes first.
  - Always state assumptions when data is ambiguous.
  - Use Python for calculations — do not guess numeric results.
  - Round all percentages to 1 decimal place.
  - Highlight the 3 most important insights in every report.

from openclaw import Agent

agent = Agent.from_config("agents/analyst-agent.yaml")

def inspect_dataset(path: str) -> str:
    return agent.run(
        f"Load the dataset at '{path}'. Report:\n"
        "1. Shape (rows × columns)\n"
        "2. Column names and dtypes\n"
        "3. Missing values per column (count + %)\n"
        "4. First 3 rows as a formatted table\n"
        "5. Basic descriptive stats for numeric columns"
    )

def detect_anomalies(path: str, column: str) -> str:
    return agent.run(
        f"In dataset '{path}', column '{column}':\n"
        "1. Calculate mean and standard deviation.\n"
        "2. Flag rows where the value is beyond 3 standard deviations.\n"
        "3. Also flag values that are statistically improbable using IQR method.\n"
        "4. Return a JSON list of anomalous row indices with their values and z-scores."
    )

def analyse_trend(path: str, date_col: str, value_col: str, period: str = "month") -> str:
    return agent.run(
        f"Dataset: '{path}'. Analyse the trend of '{value_col}' over time.\n"
        f"Group by {period}. Calculate:\n"
        "- Period-over-period growth rate (%)\n"
        "- 3-period moving average\n"
        "- Overall trend direction (up/down/flat)\n"
        "- Seasonality if detectable\n"
        "Return as JSON suitable for a line chart."
    )

import pathlib, datetime

def generate_report(path: str, output_dir: str = "./reports") -> str:
    report_md = agent.run(
        f"Perform a full analysis of '{path}'. Generate a Markdown report with:\n"
        "# Executive Summary (3 bullet points)\n"
        "## Data Overview\n"
        "## Key Metrics Table\n"
        "## Top 3 Insights\n"
        "## Anomalies Found\n"
        "## Trends and Forecast\n"
        "## Recommendations\n"
        "Use concrete numbers throughout."
    )
    output = pathlib.Path(output_dir)
    output.mkdir(parents=True, exist_ok=True)
    date_str = datetime.date.today().isoformat()
    out_path = output / f"report-{date_str}.md"
    out_path.write_text(report_md, encoding="utf-8")
    return str(out_path)

def forecast_next_periods(path: str, date_col: str, value_col: str, periods: int = 4) -> str:
    return agent.run(
        f"Using data from '{path}' (date='{date_col}', value='{value_col}'):\n"
        f"1. Fit a simple linear regression trend.\n"
        f"2. Forecast the next {periods} periods.\n"
        "3. Return JSON: {\"forecast\": [{\"period\": \"...\", \"predicted\": 0.0, "
        "\"low_95\": 0.0, \"high_95\": 0.0}], \"r_squared\": 0.0}"
    )