Full Suite for Claude Code

A complete package for Claude Code that gives it specialist agents, persistent memory, deep web search, and a disciplined workflow — all in one drop-in directory.

Drop the .claude/ folder and CLAUDE.md into your project and Claude Code automatically gets everything below. No daemons, no subprocess spawning, no configuration server. Everything is plain Markdown + a couple of POSIX/PowerShell scripts.

Agent Quality — Real-Project Tested

All 111 agents have been tested and optimized through a rigorous methodology:

Method: Each agent was tested in a 3-way comparison (original / polished / smart-applied) on real production codebases — not synthetic tasks. Tested on real production codebases.

Result: The winning variant for each agent was selected based on objective criteria: finding accuracy, evidence quality, cross-file tracing depth, and zero false positives. Agents were compared head-to-head against Claude Code's own native subagents — our agents won every comparison.

Cognitive Mode Tags: Each agent in INDEX.md carries a Mode tag (TRACE / SWEEP / KNOW) derived from these tests, indicating which cognitive approach it's best at. This helps match the right agent to the task type.

• TRACE — best at following data/logic/flow through code (bug hunting, pipeline analysis)
• SWEEP — best at systematic checklist verification (security audits, idiom reviews)
• KNOW — best at applying deep domain/framework expertise (.NET, Spring, Django)

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/

If you already have a CLAUDE.md, append this one instead of overwriting. Add tmp/, .claude/knowledge.md, and .claude/session.md to .gitignore if you prefer.

What's included

109 Specialist Agents

Domain experts spawned on demand as Task tool subagents. Each is a Markdown file with checklists, anti-patterns, and decision tables validated against n=600 evaluations. Categories include PostgreSQL, Swift, Kubernetes, React, security review, frontend, Rust, Python, and 100+ more.

Browse .claude/agents/INDEX.md for the full directory. Selection rule: most specialized wins — postgres-pro over database-optimizer, swift-pro over a generic alternative.

Only agents from .claude/agents/ may be used as Task subagent types. Built-in/standard subagents are forbidden — if no .claude/agents/ agent fits, do the work yourself directly.

Persistent Memory

Two tiers stored as plain Markdown:

# Save permanent knowledge
.claude/tools/memory.sh add gotcha "psycopg2 needs libpq-dev" --tags postgres

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

Knowledge (.claude/knowledge.md) — architecture, patterns, gotchas. Survives across sessions. Session (.claude/session.md) — plans, todos, checkpoints. Survives context compaction. Multiple instances stay isolated via -S session-name.

Deep Web Search

50+ results per query via DuckDuckGo + Brave, with topic-aware bonus sources:

.claude/tools/web_search.sh "React server components" --tech    # + HN, SO, GitHub
.claude/tools/web_search.sh "CRISPR delivery" --sci --med       # + arXiv, PubMed

Auto-installs Python deps via uv. Anti-bot bypass, smart extraction, token compression, non-English support. All public APIs — no keys needed.

Workflow

CLAUDE.md configures Claude Code with a disciplined, research-first workflow:

• Research before implementation — read reference sources, understand project structure
• Specialist delegation — spawn domain experts from .claude/agents/ as Task subagents for multi-domain tasks
• Self-verification — re-read diffs, trace control flow, run build+tests after every change
• Iteration to convergence — for high-stakes work, repeat review cycles until no new findings
• Quality over speed — methodical work produces correct code; speed produces bugs

Requirements

• Claude Code
• Python 3.9+ (dependencies auto-installed by uv)
• POSIX shell (bash/zsh) on macOS/Linux, PowerShell on Windows

License

MIT — see LICENSE.