dev++

dev++ is a compact skill set for product discovery, agentic UI validation, and safer AI-assisted software development. It helps agents turn early ideas into PRDs, user stories, assumptions, validation plans, and test scenarios before moving into code. It can then validate UI stories in a real browser with screenshots. During development, it keeps agents from jumping straight into implementation, asks for missing decisions, inventories APIs and frontend reuse points, plans tests first, and gates verification and git actions.

Translations: 简体中文 | 日本語

Development Skills

Skill	Purpose
dev-intake-approval	Clarify, inspect, find reuse, define contracts/tests, and wait for approval.
api-contract-inventory	Map frontend user flows to backend APIs, contracts, errors, and gaps.
frontend-reuse-workflow	Inventory reusable UI, prefer templates, and enforce frontend error handling.
test-and-verify	Discover and run only approved tests, builds, lint, and checks.
git-change-control	Inspect and gate commits, pushes, merges, tags, and releases.

Product Discovery Skills

Skill	Purpose
create-prd	Create a structured PRD with problem, objective, segment, value proposition, solution, and release framing.
user-stories	Break features into user stories with acceptance criteria using 3 C's and INVEST.
job-stories	Express user situations and motivations with JTBD-style job stories.
wwas	Create Why-What-Acceptance backlog items with strategic context.
test-scenarios	Turn user stories into acceptance-oriented test scenarios.
identify-assumptions-new	Map risky assumptions for a new product idea.
identify-assumptions-existing	Map risky assumptions for a feature in an existing product.
prioritize-assumptions	Prioritize assumptions by impact and risk, then suggest experiments.
brainstorm-experiments-new	Design lean startup experiments for a new product.
brainstorm-experiments-existing	Design low-effort validation experiments for an existing product feature.
opportunity-solution-tree	Map outcomes, opportunities, solutions, and experiments.
interview-script	Create customer interview scripts for discovery research.
summarize-interview	Summarize interview transcripts into JTBD-oriented insights and actions.

Agentic UI Testing Skills

Skill	Purpose
playwright-ui-automation	Drive isolated browser sessions through Playwright CLI with screenshots, console inspection, and headed/headless modes.
ui-story-validation	Validate one UI story step by step, save screenshots, capture failures, and report parseable pass/fail results.

Native Commands

After install, dev++ exposes command entries for both Codex and Claude Code.

Workflow	Codex	Claude Code
Initialize repository context	/prompts:dev-init	/dev-init
Initialize API inventory	/prompts:api-init	/api-init
Initialize frontend inventory	/prompts:frontend-init	/frontend-init
Run product discovery	/prompts:discovery-plan <idea>	/discovery-plan <idea>
Run UI story review	/prompts:ui-review [headed] [filter] [vision]	/ui-review [headed] [filter] [vision]
Plan a development request	/prompts:dev-plan <request>	/dev-plan <request>
Plan multi-agent work	/prompts:multi-agent-plan <request>	/multi-agent-plan <request>
Plan verification	/prompts:verify-plan	/verify-plan
Inspect git state	/prompts:git-check	/git-check

Install

git clone https://github.com/Fubuki233/dev-plus-plus.git
cd dev-plus-plus
scripts/install.sh

The installer copies:

• Codex skills into ${CODEX_HOME:-$HOME/.codex}/skills.
• Codex command prompts into ${CODEX_HOME:-$HOME/.codex}/prompts.
• Claude Code slash skills into ${CLAUDE_HOME:-$HOME/.claude}/skills.

Install only one side when needed:

scripts/install.sh --codex-only
scripts/install.sh --claude-only

If just is installed, the same workflows are available as stable command aliases:

just --list
just install
just install-codex
just install-claude
just validate
just ui-review-help
just init-ui-review-stories

Validate

scripts/validate.sh

or:

just validate

Documentation Convention

dev++ recommends storing generated project intelligence under docs/ai-context/. Keep the root index small, split large API and frontend inventories by domain, and load details only when needed.

Use templates/ai-context/ as a starter structure when a project has no existing convention.

For agentic UI review, store executable story YAML files under ai_review/user_stories/. Use templates/ai-review/user_stories/example.yaml as the starter format.