NESMA Workload Breakdown Workflow

Problem

When a requirement document is ready but still lacks a reviewable function-point inventory, the NESMA Workload Breakdown Workflow turns the document into a traceable estimate. It targets engineers and estimation owners who need to convert requirements into EI/EO/EQ/ILF/EIF, complexity levels, ADD/CHG/DEL change types, and finally a CSV export plus a workload estimate file.

How It Works

The skill runs through six steps and tracks progress with TodoWrite:

• Workspace preparation: creates results/<requirement-name>/ for resumable output.
• Module and function survey: a Task subagent reads the requirement and produces a survey plan, grouping candidate functions by data maintenance, business queries, interface interaction, process control, and UI interaction.
• User confirmation: the main skill summarizes modules, excluded sections, and extension suggestions, then asks whether the module structure is sound and which optional extensions to include.
• NESMA decomposition: a decomposer subagent fills function points, change types, and coefficients.
• Review: an independent reviewer subagent checks cross-functional gaps, symmetry, multi-business-line coverage, description depth, and complexity boundaries, then writes a review report.
• Export and conversion: the main skill generates a detailed CSV and runs fp2effort.py to calculate UFP → AFP → person-days.

The key design constraint is separation: workspace setup, surveying, decomposition, and review must be delegated to subagents. The main skill handles progress inference, user confirmation, and final export. This reduces confirmation bias from a single generation-and-self-check loop.

Boundaries

The skill requires a .md or .docx requirement document and a results/ folder in the project root. The default conversion factors use a telecom industry baseline, including quality, language, and team factors whose product is 0.576; for other industries, update the factors in the workload estimate file and rerun the conversion. The review loop is capped at two rounds, after which human judgment is required.