Data Analysis Skills

# Spreadsheet Analysis Skill Use this skill to analyze spreadsheet documents (Excel `.xlsx`, `.xls`, `.ods`, `.csv`) stored in Vertesia using Python (`pandas`, `openpyxl`) inside a sandbox. > **When to use this skill vs data-platform:** > - **Use this skill** for one-off analysis of uploaded files, quick exploration, or when working with Vertesia document IDs > - **Use `data_analysis` (data-platform)** when data is in a persistent data store, you need SQL queries, or want versioned/reusable analytics The goal is to: - **Discover the dataset shape first** (sheets, columns, types, ranges, missingness). - Then perform **targeted analysis** (aggregations, trends, segments) using that schema. - Use **artifacts** (`write_artifact`, `read_artifact`, `grep_artifacts`) to pass data and reuse scripts/results across steps. Important: - You **must** work with Vertesia document IDs, not local file system paths like `/Users/...`. - Whenever you need to analyze spreadsheets, you must first ident...

# Create Spreadsheet Skill Use this skill to create spreadsheets (Excel `.xlsx` or `.csv`) from data available in Vertesia or artifacts. The recommended pattern is: 1. **Stage inputs as artifacts** (CSV/JSON/text) or pull them from Vertesia documents. 2. **Write a Python script as an artifact** under `scripts/`. 3. **Execute the script via `execute_shell`** in the sandbox. 4. **Write the spreadsheet into `~/out/`** so it becomes an `out/*` artifact that can be reused or downloaded. For long-lived knowledge base objects, persist the resulting `.xlsx` artifact with `create_document(source: "artifact:out/<name>.xlsx")`. --- ## 1. Prepare input data as artifacts or documents ### 1.1 From existing documents If data already lives in Vertesia: 1. Use `search_documents` to find the right documents and get their `id`s. 2. Use `fetch_document` when you need text content for context, or: 3. Use `execute_shell` with `documents=[{"id": "<DOC_ID>", "mode": "file" | "text"}]` so the source ...

# Chart & Dashboard Creation Create visualizations using Vega-Lite via markdown code blocks. The UI renders them automatically with tooltips and interactivity. This is the entry point for charts shown inline in the conversation: one-off charts, inline or artifact-backed data, interactive exploration. Learn `learn_data_visualization` instead when the data lives in a persistent data store and the user wants a saved, reusable, SQL-backed dashboard. From here, learn `learn_vega_reference` for the Vega-Lite grammar (marks, encodings, transforms, scales, number and date formatting, color schemes), `learn_vega_reference_advanced` for faceting, statistical charts, annotations, maps and the interactive dashboard patterns (cross-filter, brush, legend toggles, input controls, multi-stage filters), and `learn_view_chart` to render a PNG of a spec and check it yourself before presenting it. Chart block formatting and artifact references are also covered in `learn_output_reference`. --- ## When...

# Mermaid Diagram Skill Use Mermaid syntax to create visual diagrams directly in your responses. Diagrams are rendered automatically when you use a `mermaid` code block. > **See also:** `output_reference` skill for general output formatting guidelines and when to use different visualization types. ## When to Use Use Mermaid diagrams to: - Illustrate workflows and processes (flowcharts) - Show system interactions (sequence diagrams) - Document data models (ER diagrams) - Explain state machines (state diagrams) - Visualize project timelines (Gantt charts) - Map out class relationships (class diagrams) - Create mind maps for brainstorming ## Syntax Wrap your Mermaid code in a fenced code block with `mermaid` as the language: ````markdown ```mermaid graph TD A[Start] --> B{Decision} B -->|Yes| C[Action] B -->|No| D[End] ``` ```` ### Export-Safe Label Formatting (PDF/DOCX) When diagrams may be exported to PDF or DOCX, prefer renderer-safe labels: - Use Mermaid line br...

# Update Spreadsheet Skill Use this skill to **update existing spreadsheet documents** stored in Vertesia: - Adjust or clean data. - Add or modify sheets. - Append new rows or computed metrics. - Regenerate summary tabs, while preserving raw data where appropriate. All updates are done via code running in the sandbox, with **scripts and data managed as artifacts**. --- ## 1. Locate and download the spreadsheet 1. Identify the document: - If the user knows the document ID, use it directly. - Otherwise, use `search_documents` (by name, type, or full_text) to find the correct spreadsheet document and get its `id`. 2. Download the file into the sandbox using `execute_shell` and the `documents` parameter: - `documents=[{"id": "<DOC_ID>", "mode": "file"}]` - `command="ls -l ~/documents"` The sandbox will contain: - `~/documents/<DOC_ID>.<ext>` (e.g. `.xlsx` or `.csv`). > Treat `~/documents` as the source of truth for the original spreadsheet during this run. --- ## 2. Auth...

# View Chart (Model Self-Check) Render a chart to PNG so **you can see it** and verify it looks correct. The user sees the chart through the chart block in your reply, not through this PNG - this is for your own visual verification. Build the chart with `learn_make_chart` first; this skill only renders the spec you already wrote. ## When to Use - You want to verify a chart looks correct before finalizing - User asks you to check/verify the chart appearance - You need visual feedback on complex visualizations - Debugging chart layout or styling issues ## How to Render Every step below runs in the sandbox through `execute_shell` — the rendering is a Python program, so there is no tool-only path to the same result. Write the script as an artifact and run it; anything written under `~/out/` syncs back as `out/*`. Use Python with `vl-convert` to render any Vega-Lite spec: ```python import vl_convert as vlc # Your Vega-Lite spec (same format as chart code blocks) spec = { "$sche...

# Excel Editing Skill Use this skill to make **targeted edits to an existing `.xlsx` artifact** — changing cell values, applying formulas, modifying formatting, adding or removing rows/columns, managing sheets. > For creating a new spreadsheet from data, see the `create_spreadsheet` skill. > For analyzing/querying spreadsheet data, see the `analyze_spreadsheet` skill. > For DOCX editing, see the `docx_editing` skill. For PPTX editing, see the `pptx_editing` skill. Unlike DOCX/PPTX, Excel uses a shared-strings table and cross-referenced XML parts that aren't amenable to direct xpath patching. The correct approach is **openpyxl in the sandbox** — it handles the data model correctly and supports formulas, formatting, and multi-sheet workbooks. --- ## Workflow ### 1. Get the .xlsx artifact ```text # From a Vertesia document: fetch_document(id: "doc_123", output_artifact: {}) # → creates docs/doc_123.xlsx # Already exists as an artifact: list_artifacts(prefix: "out/") # → out/repor...