❝ Same product brain as v1 — leaner runtime. A small always-on governance core, the full operating manual split into 11 reference files loaded on demand, and graded evals to keep behaviour honest. ❞

💜 If this skill saves you time, please give it a ⭐ — it helps other builders discover it!

Hand Claude Code a one-line idea and it will happily start coding — skipping problem validation, missing stakeholders, dependency-safe sequencing, and the security/privacy/accessibility audits that real products need.

Ultimate Idea Watcher v2 behaves like one coordinated senior product organization (strategy, design, engineering, security/privacy, QA) and takes responsibility for your idea across its whole lifecycle:

Raw Idea → Preserved Idea → Auto-Filled Brief → Evidence & Confidence Model
→ Deep Research → Product Direction → Complete Product Architecture
→ Enhancement Passes → Phase Plan → Dependency Audit
→ Phase Stage 1 → Design Verification → Phase Stage 2 → Implementation Verification
→ Cross-Phase Audit → Production-Readiness Audit → Continuous Learning → Ongoing Enhancement

This is a structural rebuild of ultimate-idea-watcher with identical capability and a much better runtime shape:

• ⚡ Lean core, loaded references. SKILL.md holds only the always-in-force governance (identity, mission, activation, clarification policy, two-stage model, response modes, start sequence, 65 non-negotiable rules, success definition). The heavy contracts and templates live in references/ and are loaded only when that part of the lifecycle is reached — leaner context, sharper behaviour.
• 🧪 Evals included. An evals/evals.json suite encodes what "good" looks like with concrete prompts and graded assertions.
• 🏷️ Versioned metadata in the frontmatter (version: 2.0.0).

👉 New here? Start with v2. Prefer the original single-file build? See the ultimate-idea-watcher repo.

• 🧷 Preserves your original idea verbatim while building a structured interpretation alongside it.
• 📝 Auto-fills a product brief (problem, users, features, business goal, security/privacy, devices, market, constraints) — no form required.
• 🔍 Separates the problem from the proposed solution instead of assuming the first idea is the best.
• 📚 Runs deep research and converts findings into decisions, each with evidence class + confidence rating and the risk if wrong.
• 🏗️ Defines the complete product before phasing: users & roles, objects, modules, features, workflows, state machines, permissions, business model, MVP.
• 🧱 Splits work into dependency-safe phases and proves coverage with a requirement traceability matrix.
• 🎨 Generates ready-to-paste prompts: a Stage 1 designer prompt (screen inventory, all states, responsive, accessibility, motion/3D with fallbacks, prototype wiring) and a Stage 2 engineering prompt (architecture, data, APIs, auth, security, jobs, testing, CI/CD, rollback).
• 🛡️ Audits continuously: gap register with severity, contradiction detection, change-impact analysis, production-readiness audit.
• ♿ Security, privacy, and accessibility are defaults — raised before you ask.

Claude Code loads skills from ~/.claude/skills/ (personal) or <project>/.claude/skills/ (project-scoped). Because v2 ships reference files, clone the whole folder so the references/ directory travels with SKILL.md.

git clone https://github.com/satishTheLegend/ultimate-idea-watcher-v2.git \
 ~/.claude/skills/ultimate-idea-watcher-v2

# from your project root
git clone https://github.com/satishTheLegend/ultimate-idea-watcher-v2.git \
 .claude/skills/ultimate-idea-watcher-v2

git submodule add https://github.com/satishTheLegend/ultimate-idea-watcher-v2.git \
 .claude/skills/ultimate-idea-watcher-v2

Restart Claude Code (or start a new session) so it discovers the skill. ✅ The references/ files are loaded on demand by the skill itself — you don't need to do anything beyond keeping them next to SKILL.md.

Claude Code reads the description in SKILL.md frontmatter and auto-activates the skill when your request matches (manual /ultimate-idea-watcher-v2 also works). It triggers when you share a product idea, ask to plan/scope/research/validate, want phased work, a screen inventory, a design system, design-before-code prompts, or want to audit/continue a plan for gaps.

The governance in SKILL.md is always in force; the skill loads the matching reference file the moment it reaches that part of the lifecycle:

"My parents run a small independent pharmacy and track expiring medicines, low stock, and reorders on paper and WhatsApp. Turn this into a real product plan."

→ Preserves the raw idea, auto-fills a brief, researches the domain, labels assumptions with confidence, separates pharmacist (operator) from owner (buyer), auto-includes security/privacy for sensitive health data, produces a dependency-ordered phase plan + traceability matrix, ends with an explicit next action.

"Here's my freelance-invoicing SaaS: sign up, create invoices, send to clients, pay via Stripe, dashboard for paid vs unpaid. Before I hand it to a dev — what am I missing on security, privacy, and edge cases?"

→ Surfaces IDOR/object-level access, Stripe webhook verification, rate limiting, retention/deletion/consent, failed/partial payments, refunds, disputes, void/edit invoice — each with severity and a remediation order.

"I'm building a habit-tracking mobile app — streaks, reminders, progress charts. I want the design fully figured out before any code. Give me the screen list and a designer prompt for phase 1, genuinely mobile-first and accessible."

→ Screen inventory, multiple states per screen, explicit accessibility, mobile as a real workflow, motion with reduced-motion fallbacks, prototype wiring with no dead controls, and a clearly demarcated designer prompt.

🧪 These three scenarios are exactly the cases encoded in evals/evals.json.

ultimate-idea-watcher-v2/
├── SKILL.md # lean always-in-force governance core
├── references/ # progressive-disclosure operating manual (loaded on demand)
│ ├── 00-organization-and-scope.md
│ ├── 01-brief-and-fields.md
│ ├── 02-evidence-and-learning.md
│ ├── 03-intake-and-research.md
│ ├── 04-product-architecture.md
│ ├── 05-enhancement.md
│ ├── 06-phases.md
│ ├── 07-stage1-design.md
│ ├── 08-stage2-implementation.md
│ ├── 09-verification-and-traceability.md
│ └── 10-output-formats.md
├── evals/
│ └── evals.json # graded scenarios (full-plan, gap-audit, stage-1 design)
├── README.md # this file
├── LICENSE # MIT
└── .gitignore

evals/evals.json encodes three representative scenarios with concrete prompts and graded assertions:

Use them as a regression bar when you edit the skill. ✅

Contributions are welcome! 🎉

1. 🍴 Fork the repository.
2. ✏️ Edit SKILL.md (governance) or the relevant references/*.md file (contracts/templates). Keep the frontmatter name/description/version accurate — description drives auto-activation.
3. 🧪 If behaviour changes, update evals/evals.json so the assertions still describe "good".
4. 🛡️ Keep additions evidence-based, dependency-safe, and security/privacy/accessibility-aware.
5. 🚀 Open a pull request describing the change and its rationale.

If Ultimate Idea Watcher v2 helped you plan, scope, or ship something, please give it a star — it genuinely helps others discover the project and motivates continued work. 🙏

Star this repo

💬 Have an idea or found a bug? Open an issue or submit a PR.

Released under the MIT License — free to use, modify, and distribute.