Plot Data

Goal

Turn data in the conversation context and user prompt into a clear, useful plot with deterministic rendering and artifact handling.

Always produce a plot that can be reproduced from saved artifacts.

Workflow

1. •

   Resolve plotting intent and data inputs.

   • •Required: at least one usable dataset (table, series, key-value list, or CSV-like text).
   • •Optional: chart type, title, axis labels, grouping, color preference.
   • •If required data is missing or malformed, ask one concise follow-up question describing exactly what is needed.
   • •If no chart type is specified, choose one by rule:
     • •time-indexed values -> line chart
     • •category comparisons -> bar chart
     • •distribution -> histogram
     • •relationship between two numeric columns -> scatter plot
2. •

   Normalize and validate data before plotting.

   • •Convert input into explicit columns and rows.
   • •Validate numeric fields for plotted measures; drop or report invalid rows.
   • •For time-series data:
     • •parse timestamps to a consistent format
     • •if the scope is within 24 hours, use hour labels on the X axis to prevent text overflow
   • •If data cleaning changes results materially (for example: dropped rows), mention it in the response.
3. •

   Render with deterministic plotting rules.

   • •Use matplotlib for chart generation.
   • •Always use uv for Python-related operations.
   • •Prefer Python standard library plus matplotlib; avoid adding dependencies unless strictly required.
   • •Use readable defaults:
     • •figure size appropriate for the data density
     • •axis labels and title when available
     • •legible tick density and rotation for long labels
     • •visible grid for quantitative charts unless user asks otherwise
4. •

   Produce deterministic artifacts.

   • •Save outputs to output/plot-data.
   • •If directory does not exist, create it using Python standard library (os).
   • •Generate a unique basename:
     • •plot_<chart-type>_<topic-slug>_<uuid8>
   • •Save:
     • •<basename>.png as the final chart image
     • •<basename>.json metadata with rendering context
   • •Metadata JSON must include:
     • •request: inferred user intent and options
     • •chart: chart type, title, axis labels, styling choices
     • •data_summary: row count, column names, dropped/cleaned rows count
     • •artifacts: absolute or workspace-relative artifact paths
     • •generated_at: ISO-8601 timestamp in UTC
5. •

   Return concise user-facing output.

   • •Provide a short summary of what was plotted.
   • •Include the artifact path(s) for PNG and metadata JSON.
   • •If assumptions were made (for example chart-type default), state them briefly.
6. •

   Handle failures explicitly.

   • •If plotting fails due to invalid data shape, report the specific mismatch.
   • •If execution fails due to transient runtime issues, retry once.
   • •Never fabricate data points to make the plot render.
7. •

   Clean up temporary files only.

   • •Remove temporary scripts and intermediate files created during plotting.
   • •Do not delete final artifacts inside output/plot-data.