# Analyze Requirements

> Create or update professional architecture documentation through a structured interview, live research, and code scanning. Use when the user asks for architecture docs, system documentation, an architecture review doc, C4 diagrams, ADRs, a threat model, or to document an existing codebase.

- Skill: `vukhanhtruong/analyze-requirements` (Agent Skill, multi-file: 114 files)
- Install (CLI): `npx skillmds add vukhanhtruong/analyze-requirements`
- Raw SKILL.md: https://api.skillmd.com/api/skills/vukhanhtruong/analyze-requirements/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- Author: vukhanhtruong (https://skillmd.com/u/vukhanhtruong)
- Updated: 2026-09-10
- Page: https://skillmd.com/skills/vukhanhtruong/analyze-requirements

---


# analyze-requirements

Produce provenance-tagged architecture documentation with interactive diagrams,
served on localhost for review.

## Hard rules

1. Every fact carries provenance: `observed` | `stated` | `researched` (with source) | `proposed`. A claim nobody verified must never render like one that was.
2. One home per fact — diagrams own topology, tables own properties, prose owns neither.
3. Unknowns render as honest absences (`Not applicable — <reason>`, `not estimated`, `Not provided`) — never placeholders, never `[TODO]`, never `0`.
4. Validation blocks rendering: `node scripts/validate.mjs` must exit 0 before the viewer is generated.
5. mattpocock files (`CONTEXT.md`, `CONTEXT-MAP.md`, `docs/adr/`) are written into their existing formats — never restructured. Invoke the `domain-modeling` skill for term work.

## Flow

1. **Detect mode**: target has source code → brownfield, else greenfield. State the detection and let the user override.
2. **Scan** (brownfield): `index_repository` if needed, then `get_architecture` — clusters seed §6 Core Components. Read `manage_adr` if present; never write it.
3. **Interview**: follow `references/interview.md`. Detect project type per `references/project-types.md`.
4. **Research**: run `workflows/research.js` per `references/research.md`. Surface dropped items before writing.
5. **Write**: model first (`references/likec4.md`), then ARCHITECTURE.md and companions (`references/writing.md`).
6. **Validate**: export the model with `node scripts/likec4-export.mjs --dir ... --out ...`, then `node scripts/validate.mjs --arch ... --model ...` (add `--mode brownfield --clusters ...` when applicable). Fix findings; re-run until clean.
7. **Render + serve**: follow `references/viewer.md`; report the URL.

## Dependency

Node ≥ 20 with npm (`npx likec4`). State this upfront; without it, stop before step 5.