lucasbiason/requirements-extractor

Requirements Extractor

Purpose

Convert unstructured input (plain text, notes, partial docs, meeting summaries) into:

• Functional Requirements (FR-xxx)
• Actors
• Flows (happy path + edge cases)
• Explicit Unknowns and Open Questions
• Assumptions (only if clearly implied)

This skill does NOT design solutions. This skill does NOT write code.

When to Use

• User provides raw text, tickets, or partial docs
• Plan Lead routes input that is "text/ideas" to this skill

Inputs Supported

• Plain text notes, existing docs (even messy)
• Tickets / Slack messages / emails
• User story fragments, problem statements

Output Files (canonical)

Write results into:

• spec/10-requirements/functional-requirements.md
• spec/99-meta/open-questions.md
• spec/99-meta/assumptions.md

If these folders don't exist, propose creating them.

Extraction Rules (hard constraints)

1. Never invent requirements. If unknown → mark UNKNOWN + add an Open Question.
2. Prefer observable behavior over internal implementation.
3. Every requirement must be testable (at least in principle).
4. Avoid UI/tech details unless the input explicitly mandates them.
5. Split combined requirements into atomic ones.

Process

Step 0 — Intake Summary

Summarize the input in 5–10 bullets: domain/subject, key capabilities, stakeholders, constraints mentioned, missing context.

Step 1 — Identify Actors

List actors (human roles, system actors) with goals, typical actions, notes/constraints.

Step 2 — Extract Functional Requirements

Create FRs with IDs (FR-001, FR-002…). Each FR must include: Title, Description (observable behavior), Primary actor, Preconditions, Success outcome, Failure/exception notes, Priority (P0/P1/P2 or UNKNOWN).

Step 3 — Group Into Capabilities

Group FRs into capability buckets (CAP-<name>: list FRs).

Step 4 — Unknowns, Ambiguities, Conflicts

Populate Open Questions (OQ-001, OQ-002…): Question, Why it matters, Suggested answer options (if appropriate), Who likely knows (role).

Step 5 — Assumptions (optional)

Only add assumptions if clearly implied by the input. Mark as A-001: <assumption>, Risk if false, How to validate.

Output Format

Use templates: functional-requirements.md with Actors, Capability Map, Requirements (FR-xxx with Description, Primary actor, Preconditions, Success outcome, Failure/exceptions, Priority). Add Traceability Notes (link to PRD items if they exist). open-questions.md and assumptions.md with the structures above.