phuetz/code-buddy/data-charts

Analyze data & render charts with pandas + matplotlib

Turn tabular data (CSV/JSON/Excel/inline) into a real chart image using the open-source pandas + matplotlib libraries. Work in visible steps and verify the image was written before reporting.

Getting the data when it's NOT in a file (e.g. "France GDP over 10 years")

When the user asks to chart something you must look up rather than read from a file, gather the numbers FIRST with your normal tools — your own knowledge, or a quick web_search — then hard-code them as inline Python lists. Do NOT fetch from an external API inside the Python script (urllib/requests): sandboxed network calls routinely time out (the 30 s command limit), which makes you give up and fall back to a much worse ASCII/text "chart". Inline numbers → matplotlib → a real PNG is fast and reliable. Only read from the network when the user explicitly gives you a data URL and it's known-fast.

Preflight (extras-tier libs)

These libs are optional and not in the default build. Always check first:

python3 - <<'PY'
try:
    import pandas, matplotlib  # noqa: F401
except ModuleNotFoundError as e:
    raise SystemExit(
        f"{e.name} not installed. Run `npm run prepare:python:extras` in cowork/ "
        "to enable the data-charts skill (pandas + matplotlib)."
    )
print("data-charts deps OK")
PY

Workflow (one step at a time)

1. Load the data with pandas (read_csv / read_json / read_excel), or

build a DataFrame from inline numbers.

1. Inspect: print df.head() and df.describe() (or the shape) so the data

is visibly understood before plotting.

1. Plot: use the non-interactive Agg backend (matplotlib.use("Agg"),

set before importing pyplot — there is no display), build the chart, add a title + axis labels, and savefig(path, dpi=144, bbox_inches="tight").

1. Verify: assert os.path.getsize(path) > 0 and print the path + size.
2. Report the saved image path and a one-line insight from the data.

Quality conventions

• Always set matplotlib.use("Agg") before import matplotlib.pyplot — the

app has no X display; the interactive backend would crash.

• Label axes and give every chart a title; use a legend when >1 series.
• Prefer tight_layout() / bbox_inches="tight" so labels aren't clipped.
• Save PNG by default (144 dpi); SVG when the user wants a vector.

Example

python3 - <<'PY'
import os
import pandas as pd
import matplotlib
matplotlib.use("Agg")            # headless — set before pyplot
import matplotlib.pyplot as plt
 
df = pd.read_csv("sales.csv")    # columns: month, revenue
print(df.head()); print(df.describe())
 
fig, ax = plt.subplots(figsize=(8, 4.5))
ax.bar(df["month"], df["revenue"], color="#3b82f6")
ax.set_title("Monthly revenue"); ax.set_xlabel("Month"); ax.set_ylabel("Revenue ($)")
fig.tight_layout()
out = "revenue.png"
fig.savefig(out, dpi=144, bbox_inches="tight")
 
assert os.path.getsize(out) > 0, "chart image is empty"
print(f"wrote {out} ({os.path.getsize(out)} bytes); peak month = "
      f"{df.loc[df['revenue'].idxmax(), 'month']}")
PY

Notes

• The chart image is an artifact — surface its path so the GUI can preview it.
• For quick stats without a chart, df.describe() / df.groupby(...).agg(...).
• To put a chart into an Excel/Word/PowerPoint file, hand the saved PNG to the

xlsx/docx/pptx create skills.