---
name: xlsx_merged_header_create_rungs
description: "Create a new spreadsheet with a real merged header cell over a range of columns or rows, so it actually looks and behaves merged when opened, not just repeated text in every cell. xlsx_merged_header_create_rungs: xlsx_merged_header_create_rungs: creates a new .xlsx workbook with a genuinely merged c"
---

# xlsx_merged_header_create_rungs (a Neruva verified skill for `xlsx`)

Create a new spreadsheet with a real merged header cell over a range of columns or rows, so it actually looks and behaves merged when opened, not just repeated text in every cell.

Verified helper code for `xlsx`, python. Entry points: `make_merged_sheet`.
Code hash sha256 `8e696386c6ce4a3b3c66397b46ad589158ce7d2bcaa69ace26108e68682e4994`, unsigned. Banked by `legacy-dev`.

## One call

Call `build(**args)` with an object matching this schema. You do not have to write code or match the helper signatures below.

```json
{
  "type": "object",
  "properties": {
    "header": {
      "type": "string",
      "description": "Text the merged cell should show"
    },
    "merge_start_row": {
      "type": "integer",
      "description": "0-indexed"
    },
    "merge_start_col": {
      "type": "integer",
      "description": "0-indexed"
    },
    "merge_end_row": {
      "type": "integer",
      "description": "0-indexed, inclusive"
    },
    "merge_end_col": {
      "type": "integer",
      "description": "0-indexed, inclusive"
    },
    "cells": {
      "type": "object",
      "description": "Map of 'row,col' (0-indexed) to the value for every other cell to set"
    },
    "output_path": {
      "type": "string",
      "description": "Where to save the .xlsx"
    }
  },
  "required": [
    "header",
    "merge_start_row",
    "merge_start_col",
    "merge_end_row",
    "merge_end_col",
    "cells",
    "output_path"
  ]
}
```

Example arguments:

```json
{
  "header": "SUMMARY",
  "merge_start_row": 0,
  "merge_start_col": 0,
  "merge_end_row": 0,
  "merge_end_col": 2,
  "cells": {
    "1,0": "Region",
    "1,1": "Rep",
    "1,2": "Total"
  },
  "output_path": "/tmp/report.xlsx"
}
```

```python
def build(header, merge_start_row, merge_start_col, merge_end_row, merge_end_col, cells, output_path):
    parsed_cells = {}
    for k, v in cells.items():
        r, c = k.split(',')
        parsed_cells[(int(r), int(c))] = v
    make_merged_sheet(header, ((merge_start_row, merge_start_col), (merge_end_row, merge_end_col)), parsed_cells, output_path)
    return {"path": output_path}

```

## How to use it

make_merged_sheet(header: str, merge_range: tuple, cells: dict, out_path: str) -> None
    Creates a new .xlsx with ONE merged header cell and writes it to out_path.

    header       the text the merged cell should show
    merge_range  ((r0, c0), (r1, c1)) -- 0-indexed, inclusive, the rectangle to merge
    cells        {(r, c): value} for every other cell to set, also 0-indexed
    out_path     where to save the workbook

    IMPORTANT: never write to a cell inside merge_range other than (r0, c0) -- openpyxl
    raises AttributeError on that, because every other cell in a merged range becomes a
    read-only MergedCell object once merged.

Example:
    make_merged_sheet("SUMMARY", ((0, 0), (0, 2)),
                      {(1, 0): "Region", (1, 1): "Rep", (1, 2): "Total"},
                      "/tmp/report.xlsx")


## Evidence (corpus named)

- held-out 3 of 5 specs, agent-executed (no separate model, $0): 3/3 passed a gated two-sided-contract checker (variants 2/2 accepted, spoilers 5/5 rejected on the named clause, cross-spec 12/12 rejected)
- cold naive creation (repeat the header text into every cell of the range instead of a real merge) failed 5/5, always on the same clause: no real merged range was ever created.
- one-call schema flattened after live testing with a local Qwen3-0.6B: the original schema used a nested list-of-pairs for `cells` and the model never produced it correctly (3/3 malformed or unparseable); a flat JSON object keyed 'row,col' fixed the JSON-structure problem completely (3/3 clean, well-typed JSON). Remaining failures on that same model are a genuine reasoning limit, not a schema defect: it reliably sets merge_end_row equal to merge_end_col regardless of instruction, even when told explicitly not to touch other rows -- reported honestly rather than tuned away.

## Files

- `scripts/skill.py`: the verified code (GET /v1/commons/rungs/rec_c8bb1b1f87124918b6c8e997af958205/code)
- verify: POST /v1/commons/verify {"id": "rec_c8bb1b1f87124918b6c8e997af958205"}