Business Requirements Capture

Business Requirements Capture

Generate structured Business Requirements Documents from exploration session captures.

Usage

python ./scripts/execute.py \
  --input <file> [<file2>...] \
  --mode <brd|rules|constraints> \
  --output <output_file.md>

Modes:

• brd (default): Full Business Requirements Document — functional/non-functional requirements, business rules, constraints, assumptions, success measures.
• rules: Business rules ledger only — decision logic, policy constraints, workflow conditions.
• constraints: Constraints and assumptions register only.

Flags:

• --input PATH [PATH ...] : Session brief, problem framing, BRD draft, or prior captures
• --output PATH : Destination file (default: exploration/captures/brd-draft.md)
• --mode MODE : Output scope (default: brd)

Interactive Co-Authoring Workflow

When invoked interactively, follow this 3-stage pattern. Do not dump a full document at once.

Stage 1: Context Gathering

Ask all three questions in a single message before reading or generating anything:

1. Input: Resolve input in this priority order — do not scan the filesystem before asking:
   • If the user passed a problem statement or context inline — use that as primary input (mark as [UNCONFIRMED] source)
   • If input files were named — read them after this question; note any that are missing
   • If neither exists — halt here and ask for input before proceeding (Check exploration/ only if no inline input and no files named.)
2. Output scope: Which mode fits your need?
   • brd — full Business Requirements Document (functional + non-functional requirements, business rules, constraints, assumptions, success measures)
   • rules — business rules ledger only (decision logic, policy constraints, workflow conditions)
   • constraints — constraints and assumptions register only
3. Emphasis: Any specific area, stakeholder concern, or hard constraint that must be front and centre?

After the user responds: read each input file they identify. Note any that are missing or incomplete.

Stage 2: Section-by-Section Refinement

Build iteratively — do not present a massive markdown dump.

[CONFIRMED] / [UNCONFIRMED] decision rule (apply throughout this stage):

• Mark [CONFIRMED] only when: the user explicitly stated it, OR it appears clearly in an input file.
• Mark [UNCONFIRMED] when: you inferred it, extrapolated it, or the source is ambiguous.
• When in doubt, use [UNCONFIRMED] and surface it for human sign-off.

1. Brainstorm & Curate: Based on the input files, present a numbered list of proposed requirements or rules (lightweight — title + one sentence each). Ask: "Which of these should we keep, cut, or merge? Anything missing?"
2. Draft section by section: Present all approved items as drafts in a single message — one 2–4 sentence block per item, [CONFIRMED]/[UNCONFIRMED] marked. Ask: "Accurate? Flag any item to revise before we move on." Apply all edits in one pass, then proceed.
3. Consolidate gaps: Collect all [UNCONFIRMED] items and any [NEEDS HUMAN INPUT] markers into a single ## Consolidated Gaps section at the end.

Stage 3: Reader Testing (Validation)

Once all sections are drafted:

1. Select the first reader using this heuristic — state your choice and one sentence of reasoning:
   • Technical constraints, data flows, system behavior → engineer or tech lead
   • Policy, compliance, approval gates → PM or auditor
   • Cost, timeline, stakeholder sign-off → executive sponsor
   • When multiple apply → pick the reader with the most to lose from an ambiguity in this specific document
2. Predict exactly 3 questions that reader would ask after reading this document that it does not yet answer. Make questions specific to this document's content (not generic).
3. Report: "If [reader type] reads this, they'll ask: [Q1], [Q2], [Q3]. Should we answer these inline or add them to ## Consolidated Gaps?"
4. Apply whatever the user decides.

Anti-Hallucination Rules

• Do NOT invent requirements not present in the source captures.
• Apply [CONFIRMED]/[UNCONFIRMED] markers per the Stage 2 decision rule — do not skip marking.
• Do NOT make architectural decisions — capture only what was said or decided.
• Do NOT proceed without input — inline problem statements are valid input, but a BRD generated with neither files nor inline context is pure hallucination.