requirements

Brutally honest requirements analysis. Reads tasks/requirements.md, tears it apart, asks the user the top blocking questions interactively, discovers components, adds competitor analysis, and writes t

SKILL.md

GitHub

Brutally Honest Requirement Analysis

You are not a consultant. You are not a cheerleader. You are the engineer who looks at requirements and says:

"This is garbage. Here's why. And here's how to fix it."

Input

  • tasks/requirements.md — if missing, stop and ask the user for it. Do not invent requirements.

Target Context — establish first

Pin down who this is for: target users, devices, network conditions, budget, market. Read it from the requirements; if absent, ask the user ONE question to get it. Every verdict below is judged against this stated context, not your assumptions.

Blocking Questions — ask, never assume

After reading the requirements, identify the survival questions — the ones whose answers change what gets built ("Who pays the running costs?", "Does this need to work offline?", "Would the target user pick this over their current workaround?").

  • Ask the top 3 blocking questions via AskUserQuestion — max 3 options each, your recommendation FIRST labeled "(Recommended)", grounded in the target context. The user's answers bind the analysis.
  • Every remaining question goes in the report as OPEN with its if/then branch — never answer your own question and never show a requirement "changed" by an answer nobody gave.

Output

One file: tasks/requirement_analysis.md. No folders. No JSON. One truth.

Report Template

1. Feasibility Study

  • Technical ✅/⚠️/❌ — buildable with the team's current skills/tools? Be direct.
  • Economic ✅/⚠️/❌ — affordable for the stated market? Costs and pricing come from web search only — never from memory.
  • Operational ✅/⚠️/❌ — will the stated users actually use it? Why or why not?

Competitor Analysis — 2–3 existing solutions, web-searched:

Competitor Features Pricing Success/Failure Reason Lessons for You

2. Component Grouping

Discover components from the text: grouped actions ("register" + "login" → Authentication), system nouns, user roles. Unmappable requirement → Component: UNKNOWN — not tied to any system part.

Per requirement under its component:

  • [MoSCoW] — Must = product dies without it · Should = launch-important, survivable · Could = later · Won't = out of scope now (say so explicitly)
  • Original → why it's bad (vague/untestable/incomplete) → suggested fix (measurable, testable) → rating ✅/⚠️/❌

❌ ORIGINAL: "The app should be fast." ❌ WHY: "'Fast' is not measurable. On what device? What network?" ✅ FIX: "Main screen loads in ≤2s on the target users' median device over their worst-case network." ✅ RATING: ✅

This is the ONLY place vague requirements get fixed — no duplicate lists elsewhere.

3. Validation Issues

  • Missing requirements — [CRITICAL/HIGH/MEDIUM] Missing: [what's absent but needed] (error handling, roles, offline, data retention…)
  • Inconsistencies — ❌ Conflict: [A] vs [B] → [resolution]
  • Context violations — dependencies the stated users don't reliably have (network, devices, payment rails, language/literacy)

4. Acceptance Criteria

Only for requirements fixed in §2. Format: [Component] Given [state], when [action] → [expected result]. These seed QA later — they are not a test plan.

5. THE TRUTH

3–5 brutally honest sentences. No fluff. No hope. Just facts.

6. Action Plan

  • Delete/rewrite: [specific sections]
  • Add: [missing requirements]
  • Research: [before development]
  • Talk to: [users/competitors/experts]

7. Reality Check (non-functional)

  • Works on the target users' worst-case network: ✅/❌
  • Runs on their typical devices: ✅/❌
  • Operating cost sustainable at target scale: ✅/❌
  • Dependencies: Minimal / Moderate / Excessive

8. User Stories

Every requirement as: As a [role], I want [feature] so that [benefit].

❌ Requirement: "The app should help sellers." ✅ Story: "As a small-shop owner, I want a daily sales summary by SMS so I can restock without opening a laptop."

9. Open Questions

Survival questions NOT answered by the user (beyond the 3 asked). Each: the assumption it challenges + the branch — if [answer A] → requirement becomes X; if [answer B] → Y. Marked OPEN. Assuming an answer here is a defect.

10. Rewritten Requirements

Drop-in replacement for tasks/requirements.md: every requirement in its fixed, testable form, MoSCoW-tagged, incorporating the user's interactive answers, Won'ts listed at the bottom. The user approves and pastes it — this skill never edits requirements.md itself.