Plugin Claude Code — workflow de développement complet
npx claudepluginhub vendeesign/codebloomSkills, agents and rituals for Claude Code — plan, verify, ship. No rush, no slop, no drift.
Claude Code does the work. Codebloom teaches it to do it right.
A Claude Code plugin that turns your AI agent into a structured teammate: it explains before coding, verifies before shipping, and documents before pushing.
You use Claude Code. It's powerful. But:
Codebloom fixes this with automatic guardrails that guide Claude at every step.
/plugin marketplace add vendeesign/codebloom
/plugin install codebloom@codebloom
That's it. Then in any project:
/setup ← Configure the project (once)
/hello ← Start of session
(work) ← Guardrails kick in automatically
/push ← End of session (docs + commit + push)
You: "Add a share button"
Claude: *modifies 8 files, adds a library, refactors the header, breaks responsive*
You: "... I just wanted a button"
You: "Add a share button"
Claude: "I'll add a <button> in ShareSection.jsx
with a handler calling the Web Share API.
1 file changed, no new dependency. Good to go?"
You: "Go"
Claude: *implements, tests, auto-reviews, ready to push*
Skills are guardrails that Claude loads automatically when the context matches — no command needed.
| Skill | What it does |
|---|---|
intent-check | Classify intent before acting — distinguish questions, feelings, instructions |
code-quality | Enforces STOP → plan → approval → code. Karpathy principles. |
ui-design | Prevents "AI slop" — distinctive interfaces, not generic cyan/glassmorphism. WCAG AA accessibility built-in. |
security | OWASP Top 10 prevention, STRIDE threat modeling, data classification |
backend | Backend/fullstack patterns — REST API, 3-layer architecture, auth, DB, error handling |
mobile | Mobile patterns — Android (Kotlin/Compose), iOS (SwiftUI), Flutter, React Native/Expo |
context-guard | Saves tokens — targeted reads, anti-cascade, 70% threshold |
dependency-check | Challenges every new dependency: native possible? size? alternative? |
discovery | Deep discovery when requirements are vague or complex |
git-conventions | Conventional commits, branch naming, destructive operation guards |
doc-sync | Documentation reminder after significant changes |
demo-page | HTML preview page to validate a design before coding |
wp-pack | Auto WordPress packaging — detect, version bump, zip, verify |
create-skill | Create new skills through a guided process (mental tests, bundling, 500-line limit) |
create-agent | Create new subagents through a guided process |
Agents that work in isolated context — their output doesn't pollute your conversation.
| Agent | Role | Triggers... |
|---|---|---|
reviewer | 3-perspective review: quality, security, footprint — false positive list, confidence per issue, CLAUDE.md double-check | After every code change |
tester | Generates and runs tests, QA per-page checklist, adversarial testing | After every code change |
security-auditor | Weighted security scoring (0-100), audits deps, secrets, configs | When code touches sensitive areas |
researcher | Technical research before implementation | Before using an unknown lib/API |
auditor | Full health audit (9 axes) | On demand |
doc-writer | Syncs documentation | After significant changes |
git-historian | History analysis, hotspots, blame | On demand |
A permanent status bar at the bottom of Claude Code:
Model: Opus 4.6 | /home/user/project | Session: 1h23 | Cost: $1.45
Context: 35% | 70k/200k tokens | 42k consumed
5h: ▓▓▓▓▓▓▓▓▓▓░░░░░░░░░░ 48% | Reset 2h13 | Weekly: ▓▓▓░░░░░░░ 28%
Model, duration, cost, context, rate limits — everything at a glance.
Claude notifies you when it's done or waiting for your input:
| Platform | Audio | Visual |
|---|---|---|
| Windows | SystemSounds | Orange taskbar flash |
| macOS | System sound | Native notification |
| Linux | paplay | notify-send |
Zero configuration — enabled automatically on install.
/hello starts a watcher on your screenshots folder. Take a screenshot, the path is copied to clipboard — paste it into Claude Code to send the image.