Full Suite for Claude Code

A complete agent + workflow package for Claude Code. Drop the .claude/ directory and CLAUDE.md into your project and Claude Code automatically gets:

• 109 specialized agents — domain experts loaded as needed (PostgreSQL, Swift, Kubernetes, security review, frontend, etc.)
• Single-model workflow — disciplined planning, research-before-implementation, self-verification, iteration to convergence
• Persistent memory — knowledge that survives across sessions; per-task session state with checkpoints
• Deep web search — 50+ results per query with topic-aware bonus sources (arXiv, PubMed, Hacker News, Stack Overflow, …)

Everything is plain Markdown + a couple of POSIX/PowerShell scripts. No daemons, no subprocess spawning, no configuration server. The agents are checklists Claude reads into its own context — not separate processes.

Quick Start

git clone https://github.com/itohnobue/full-suite-claude-code
cp -R full-suite-claude-code/.claude /path/to/your/project/
cp full-suite-claude-code/CLAUDE.md /path/to/your/project/

Or, if you already have a project CLAUDE.md, append the contents of this CLAUDE.md to it instead of overwriting.

That's it. Open your project with Claude Code and the workflow activates automatically.

Tip: add tmp/, .claude/knowledge.md, and .claude/session.md to your project's .gitignore if you'd rather not commit memory files.

Agents

Browse .claude/agents/INDEX.md for the full directory grouped by domain. Quick reference:

Category	Count	Examples
Language Implementation	22	python-pro, golang-pro, rust-pro, typescript-pro, c-pro, cpp-pro, java-pro, ruby-pro, scala-pro
Web Frameworks	10	react-pro, nextjs-pro, vue-pro, django-pro, fastapi-pro, rails-pro, spring-boot-pro
Architecture & Design	9	backend-architect, api-designer, microservices-architect, event-sourcing-architect
DevOps & Infrastructure	11	devops-engineer, kubernetes-architect, cloud-architect, terraform-pro, sre-engineer
Security	6	security-reviewer, penetration-tester, threat-modeling-pro, backend-security-coder
Database	5	postgres-pro, sql-pro, database-architect, database-optimizer, database-reviewer
Testing & Quality	5	code-reviewer, tdd-guide, test-automator, qa-pro, e2e-runner
AI & ML	5	ai-engineer, ml-engineer, mlops-engineer, prompt-engineer, llm-architect
Frontend & Mobile	5	frontend-developer, ios-pro, ui-designer, ux-designer, mobile-developer
Documentation	7	documentation-pro, technical-writer, docs-architect, tutorial-engineer
Incident & Troubleshooting	4	incident-responder, debugger, devops-troubleshooter, devops-incident-responder
Specialized	20	build-engineer, cli-developer, product-manager, web-searcher, agent-organizer, …

How they're used. When Claude tackles a non-trivial subtask, it reads the matching agent's .md file (e.g. .claude/agents/postgres-pro.md), applies the checklists/anti-patterns to the current work, then discards the instructions. One agent at a time, fresh re-read every time. Hybrid tasks split into subtasks, each with the appropriate agent.

Selection rule. Most specialized wins — postgres-pro over database-optimizer, swift-pro over a generic alternative.

Memory System

Two tiers, both stored as plain Markdown so they're trivially diffable and version-controllable:

• Knowledge (.claude/knowledge.md) — permanent. Patterns, gotchas, architecture notes, design decisions. Survives across sessions.
• Session (.claude/session.md) — temporary. Current task: plan, todos, progress, checkpoints, blockers.

# Save a discovery
.claude/tools/memory.sh add gotcha "psycopg2 needs libpq-dev on Ubuntu" --tags postgres,ubuntu

# Recall context before starting work
.claude/tools/memory.sh context "postgres connection"

# Track session progress
.claude/tools/memory.sh session add todo "Implement auth middleware" --status pending

# Checkpoint (after every significant step)
.claude/tools/memory.sh session add context "CHECKPOINT: auth middleware | DONE: route+model | NEXT: tests"

Multiple Claude Code instances stay separated via -S session-name or the MEMORY_SESSION env var.

Web Search

Deep web search via web_search.sh — DuckDuckGo + Brave fallback, anti-bot handling, smart content extraction, sentence-level BM25 compression, cross-page deduplication, and topic-aware bonus sources:

.claude/tools/web_search.sh "React server components best practices" --tech
.claude/tools/web_search.sh "CRISPR delivery methods" --sci --med
.claude/tools/web_search.sh "Kalman filter implementations" --sci

Flag	Adds sources
--tech	Hacker News, Stack Overflow, Dev.to, GitHub
--sci	arXiv, OpenAlex
--med	PubMed, Europe PMC, OpenAlex

CLAUDE.md instructs Claude to route ALL internet research through this tool — no rogue curl, no built-in WebSearch/WebFetch — so results are consistent and rate-limit-friendly.

Dependencies are auto-installed by uv on first run; no manual setup needed.

Workflow Model

The included CLAUDE.md configures Claude Code as a single model handling the task end-to-end — planning, research, implementation, verification, delivery. There is no spawning, no parallel agent batches.

Key principles:

• Quality over speed. No deadlines. Slow, methodical work produces correct code; speed produces bugs.
• Research before implementation. Read the actual reference source for the equivalent feature before designing your own. Skim project structure to understand layout before deep work.
• Self-verification. After every non-trivial change: re-read the diff, mentally trace control flow, run build + tests, address root causes (not symptoms), reject speculation, verify framework semantics.
• Iteration to convergence for high-stakes work — security audits, production changes, complex refactors. Two consecutive empty review passes = done. Default cap 3 (use 5 for high-stakes).
• Phase-by-phase delivery. For multi-step tasks: decompose into ordered phases, brief the user, execute one at a time, never silently skip a planned phase.

Full details in CLAUDE.md.

Requirements

• Claude Code — the host CLI.
• Python 3.9+ for the memory and web-search tools. Dependencies are managed by uv and auto-installed on first run.
• POSIX shell (bash/zsh) on macOS/Linux, PowerShell or cmd on Windows. Both wrappers are included.

No API keys or services required for the core suite. The web search tool uses public endpoints (DuckDuckGo, Brave, arXiv, etc.) — no key needed for typical use.

License

MIT — see LICENSE.