Download the PHP package mixvoip/claude-boost without Composer
On this page you can find all versions of the php package mixvoip/claude-boost. It is possible to download/install these versions without Composer. Possible dependencies are resolved automatically.
Download mixvoip/claude-boost
More information about mixvoip/claude-boost
Files in mixvoip/claude-boost
Package claude-boost
Short Description One file. Zero commands. Makes Claude smart about your codebase. Any language. Any framework.
License MIT
FAQ
Example:
In general, it is recommended to use always a project to download your libraries. In an application normally there is more than one library needed.
- Some hosting areas are not accessible by a terminal or SSH. Then it is not possible to use Composer.
- To use Composer is sometimes complicated. Especially for beginners.
- Composer needs much resources. Sometimes they are not available on a simple webspace.
- If you are using private repositories you don't need to share your credentials. You can set up everything on our site and then you provide a simple download link to your team member.
- Simplify your Composer build process. Use our own command line tool to download the vendor folder as binary. This makes your build process faster and you don't need to expose your credentials for private repositories.
Informations about the package claude-boost
Claude Boost
One file. Zero commands. Makes Claude smart about your codebase.
Drop one folder into any project — PHP, JavaScript, Python, Go, Rust, Ruby, Java, anything — and Claude becomes a senior developer who knows your entire codebase. It reads your code, builds a registry, detects duplicates, learns your conventions, sets up safety guards, and writes a lean CLAUDE.md that persists across every session.
Install (Any Project)
That's it. Claude handles everything interactively.
Laravel Projects
What Claude Does
When Claude reads learn.md, it runs a 12-step interactive setup:
- Checks for previous progress — resumes if interrupted, or enters Refresh Mode if already set up
- Discovers your stack — languages, frameworks, databases, testing tools
- Asks you questions — permission level, features to enable
- Drafts CLAUDE.md early — safety net in case of interruption
- Deep scans your codebase — language-aware scanning (not just file listing)
- Builds a registry — every class, function, route, model cataloged in grouped markdown
- Detects duplicates — synonym-aware comparison (30+ synonym groups)
- Learns your conventions — from your actual code, not imposed rules
- Installs safety hook — blocks destructive commands via shell hook
- Maps dependencies — traces who depends on what
- Creates skills, architecture docs, git standards — module docs, architecture map, branch/commit rules
- Finalizes lean CLAUDE.md — project essentials with pointers to deep docs
- Final summary — reports what was created and next steps
If interrupted at any point, just say "continue" or re-paste learn.md. Claude reads learn-progress.json and picks up exactly where it left off.
Already set up? Just run the same command again. Claude enters Refresh Mode — no questions, no setup, just re-scans and updates.
What You Get
Why This Works
| Problem | How This Solves It |
|---|---|
Claude creates formatCurrency() when convertMoney() exists |
Registry + synonym-aware duplicate detection |
| Claude forgets your architecture every session | Lean CLAUDE.md loaded every session, deep files read on-demand |
| You re-explain patterns and conventions | Guidelines and skills persist across sessions |
| Claude doesn't know your domain rules | Domain rules are in CLAUDE.md |
| Works only for one language | Works for any language — Claude reads any code |
Updating
When your codebase changes significantly:
Claude detects your existing setup and enters Refresh Mode automatically — no questions, no confirmation. It skips setup, conventions, safety hooks, and dependencies (already done), and only re-runs what matters: scan → registry → duplicates → skills/architecture → CLAUDE.md. Fast and silent.
Laravel: After Package Upgrade
This refreshes learn.md, hooks, and templates to the latest version. Then run the learn command above — Claude will use Refresh Mode automatically.
Uninstalling
To cleanly remove Claude Boost from your project:
Claude will:
- Inventory all Claude Boost files and classify them (ours vs yours)
- Ask what you'd like to keep (registry, architecture, skills, etc.)
- Back up everything before making changes
- Restore your pre-boost CLAUDE.md and settings.json from git history
- Remove only Claude Boost artifacts
- Show you the full diff to review before you commit
For Laravel projects, also run composer remove mixvoip/claude-boost after.
Laravel Commands
The Composer package provides three commands:
| Command | What It Does |
|---|---|
claude:init |
Scaffolds .claude/ directory, installs learn.md, hooks, and templates |
claude:doctor |
Health check — verifies setup, hooks, registry, learning progress |
claude:update |
Refreshes learn.md and hooks after a package upgrade |
Everything else is handled by Claude reading learn.md.
Safety
Guard hook (preToolUse.sh) blocks destructive commands in real-time:
DROP DATABASE,TRUNCATE TABLE,DELETEwithout WHERErm -rfon critical directoriesgit push --forceto protected brancheschmod 777, direct.envmanipulation- Production migrations,
curl | bash
The hook is pure bash — no PHP runtime needed. Works for any language.
Agent Pipeline — Plan, Develop, Review
Claude Boost includes 3 autonomous agents that turn your ticket workflow into a CI/CD-like pipeline:
| Agent | What It Does | Mode |
|---|---|---|
Planner (plan.md) |
Interviews you, scans codebase, creates structured tickets | Interactive |
Developer (develop.md) |
Picks up tickets, plans, uses parallel sub-agents, opens PRs/MRs | Autonomous |
Reviewer (review.md) |
4 parallel specialist reviews (quality + performance + security + breakage), batch review with inline comments | Autonomous |
Quick Start
Connecting Your Ticket Tool
The agents need a CLI tool to create tickets and manage PRs/MRs:
| Platform | CLI | Setup |
|---|---|---|
| GitHub | gh |
brew install gh && gh auth login |
| GitLab | glab |
brew install glab && glab auth login |
| Linear | Linear MCP | Add via Claude Code MCP settings |
| Jira | Jira MCP | Add via Claude Code MCP settings |
The agents auto-detect your platform from .claude/settings.json or directory structure (.github/ vs .gitlab-ci.yml).
How It Works for Teams
- You + Planner discuss the feature → Planner creates a structured ticket with file paths, acceptance criteria, and test plan
- Developer picks up the ticket → breaks it into sub-tasks → launches parallel agents → opens PR/MR
- Reviewer picks up the PR/MR → dispatches 4 specialist reviewers → posts inline comments as a batch review → approves or returns with specific fixes
- You merge — the boring part is automated
Returned tickets go back to the Developer (max 2 returns, then blocked for human intervention). See .claude/init/AGENTS.md for full documentation.
Self-Maintaining
CLAUDE.md instructs Claude to keep everything updated:
- New code created -> Claude updates registry.md
- Module changed -> Claude updates the skill file
- Major feature lands -> Claude updates CLAUDE.md
You don't maintain these files manually. Claude does it during normal development.
Publishing Your .claude Folder
Once Claude Boost has learned your project, you can publish the generated .claude folder so your entire team benefits — every developer gets the same registry, conventions, safety hooks, and project brain from the first session.
Commit to Your Project
The simplest approach — just commit the .claude folder to your project repo:
Sensitive files like learn-progress.json, settings.local.json, and logs/ are already gitignored. Everything else — registry, guidelines, skills, hooks, and CLAUDE.md — is safe and meant to be shared.
What Gets Published
| File | Shared? | Why |
|---|---|---|
CLAUDE.md |
Yes | Project essentials — every developer's Claude reads this |
registry.md |
Yes | Codebase catalog — prevents duplicate code |
architecture.md |
Yes | Module map & data flow — shared knowledge |
guidelines.md |
Yes | Conventions — consistent code style |
skills/ |
Yes | Module docs — shared knowledge |
hooks/preToolUse.sh |
Yes | Safety guard — team-wide guardrails |
settings.json |
Yes | Hook registration — auto-activates for the team |
learn-progress.json |
No | Gitignored — per-user tracking |
settings.local.json |
No | Gitignored — per-user permissions |
logs/ |
No | Gitignored — per-user guard logs |
Team Workflow
- One developer runs Claude Boost to learn the project
- Commit the
.claudefolder to the repo - Every team member gets the full context on
git pull - Claude reads CLAUDE.md automatically — no setup needed for new developers
- When the codebase evolves, re-run the learning step and commit the updates
This turns Claude from a generic assistant into a team-wide senior developer who knows your entire codebase — and stays in sync.
Why Claude Boost?
Most approaches to enhancing Claude Code rely on plugins, background services, or additional AI calls to give Claude context about your codebase. This introduces overhead that works against you:
- Extra token consumption — AI-powered compression and summarization tools make additional API calls on every session, tool use, or prompt. Those tokens add up fast, especially on large projects.
- Runtime dependencies — background daemons, vector databases, and additional runtimes add infrastructure overhead for what should be a zero-friction experience.
- Plugin system lock-in — if the plugin API changes or your environment doesn't support it, the tool breaks. Your project context shouldn't depend on a third-party lifecycle.
- Lossy context — AI-generated summaries lose detail. A compressed memory of your codebase is never as useful as a structured, complete registry.
The Claude Boost Approach
Claude already knows how to read files — it does it extremely well. Instead of building middleware that summarizes your code for Claude, Claude Boost lets Claude read structured context directly:
| Aspect | Plugin-Based Approach | Claude Boost |
|---|---|---|
| Architecture | Background services, vector DBs, AI compression | Plain files — JSON and Markdown |
| Token cost | Extra API calls per session/action | Zero additional tokens — Claude reads local files |
| Dependencies | Additional runtimes, databases, HTTP servers | None — just Claude CLI |
| Context quality | AI-generated summaries (lossy) | Structured registry — every class, function, route cataloged |
| Portability | Tied to plugin system | Drop a folder into any project, done |
| Transparency | Compressed context you can't easily inspect | Human-readable files you can review and version-control |
A registry, guidelines, and skills — all in plain files that cost zero extra tokens, survive across every session, and work with any language.
The best tools work with the system, not around it.
Requirements
Any project: Claude CLI installed. That's it.
Laravel package: PHP 8.1+, Laravel 10+, jq, git
Sponsor
Built in partnership with Mixvoip. Thanks for supporting open-source development.
Contributing
Found a bug or have an idea? Open an issue or submit a pull request.
License
MIT License
One file. Zero commands. Makes Claude smart about your codebase.
