| name | vibe-project-master-plan |
| description | Builds a complete, reviewable project plan for a vibe-coded app in July 2026, including user journeys, cold start, account/auth, model/provider readiness, architecture, milestones, tests, launch, and rollback gates. Use when a user asks for a bulletproof plan before or during fast AI-assisted product building. NOT for writing implementation code, generic task lists, or visual design critique without a product plan. |
| license | Apache-2.0 |
| allowed-tools | Read,Write,Edit,Bash,Grep,Glob,WebSearch,WebFetch |
| metadata | {"category":"Product & Planning","tags":["vibe-coding","project-planning","product-requirements","launch-readiness","port-daddy"],"provenance":{"kind":"first-party","owners":["port-daddy"]},"pairs-with":[{"skill":"product-reality-reviewer","reason":"Reviews the plan for user-need, cold-start, and product-risk gaps."},{"skill":"developer-surface-strategist","reason":"Turns plan requirements into SDK, CLI, MCP, GUI, and automation surface choices."},{"skill":"skill-architect","reason":"Keeps the planning workflow reusable and testable."}],"io-contract":{"kind":"deliverable","consumes":["[Truncated]","[Truncated]"],"produces":["[Truncated]","[Truncated]","[Truncated]"]}} |
Vibe Project Master Plan
Create a project plan that survives fast AI-assisted building: clear enough for agents to execute, skeptical enough to
catch missing product assumptions, and concrete enough to turn into PR-sized slices.
Use This For
- Planning a new app, tool, integration, SDK, or agent workflow before a vibe-coding sprint.
- Hardening an existing loose plan into a buildable sequence with acceptance gates.
- Converting "I want to build X" into user journeys, architecture, data contracts, tests, release steps, and rollback.
- Making Port Daddy or other multi-agent work legible before agents start cutting code.
Do Not Use This For
- Pure brainstorming with no intent to build.
- Code review after implementation; use a code review or product reality skill instead.
- Design polish, copywriting, or visual QA without a plan artifact.
Process
flowchart TD
A[Capture product intent] --> B[Name users and jobs]
B --> C[Map first-run and account path]
C --> D[Choose surfaces and architecture]
D --> E[Define data, integrations, and trust boundaries]
E --> F[Slice build into PR-sized milestones]
F --> G[Attach tests, evals, proof, and rollback]
G --> H[Score completeness and list gaps]
- State the product promise in one sentence, then name the non-goals.
- Define primary users, their jobs, and the first three moments after they arrive cold.
- Specify account creation, auth, secrets, model/provider readiness, and what happens if the user lacks paid AI plans.
- Choose the product surfaces: GUI, CLI, SDK, MCP, background agents, webhooks, docs, support, and telemetry.
- Define data objects, trust boundaries, permissions, privacy, retention, and cost controls.
- Break the work into PR-sized slices with acceptance tests, visual or transcript proof, and rollback.
- Run
scripts/plan_score.mjs on a JSON manifest and add its gaps to the plan.
Output Contract
Produce a plan with these sections:
Product Promise: target user, job, payoff, and explicit non-goals.Cold Start: first-run path, account path, empty states, provider readiness, and fallback mode.User Journeys: happy path, failed path, recovery path, and repeat-use loop.Architecture: surfaces, data model, integrations, auth, permissions, hosting, and observability.Agent Plan: which agents run, what they can touch, how they communicate, and how humans interrupt them.Build Slices: ordered milestones with acceptance gates, tests, proof artifacts, and rollback.Launch Readiness: docs, onboarding, pricing/cost, support, security, telemetry, and post-launch review.
Use scripts/plan_score.mjs to produce a machine-checkable plan-scorecard.
Anti-Patterns
Feature List Masquerading As A Plan
Novice: "We have ten features, so the plan is complete."
Expert: A complete plan explains who arrives, how they start, what must be true before they trust the product, what fails, and how work is proven.
Timeline: July 2026 AI-built products fail less from missing features and more from missing trust, onboarding, provider, and rollback paths.
Detection: The plan has features but no first-run state, account path, fallback mode, or acceptance artifacts.
Paid-Plan Assumption
Novice: "Users will have Claude Max, OpenAI Pro, or the same local setup I have."
Expert: Cold start must work for users with no paid model account, missing tokens, disabled MCP, or enterprise restrictions.
Timeline: By 2026, model/provider access is fragmented across consumer, team, enterprise, local, and routed-provider accounts.
Detection: The plan says "connect your AI" but omits degraded mode, token UX, provider deeplinks, or mock/demo data.
Agent Magic Without Proof
Novice: "An agent will do it."
Expert: Every agent action needs scope, trigger, input, permission, progress, receipt, test, and undo.
Timeline: Multi-agent systems in 2026 need operator control surfaces and transcripted proof, not only chat promises.
Detection: Agents appear in the plan without claims, event logs, human gates, or rollback.
References
| File | Load When |
|---|
references/complete-plan.md | Need the required sections and examples of what "complete" means. |
references/bulletproofing-gates.md | Need review gates for cold start, account setup, provider fallback, tests, and launch readiness. |
examples/expected-output.md | Need a finished example plan and scorecard shape. |
templates/output-template.md | Need a reusable project-plan template. |
schemas/project-plan.schema.json | Need the JSON input contract for the scorer. |
scripts/plan_score.mjs | Need deterministic plan completeness scoring. |
agents/openai.yaml | Need a subagent descriptor for delegated planning. |
Skill Bundle Index
root
agents/
examples/
references/
schemas/
scripts/
templates/
