claude-code-patterns

Development harness extracted from Claude Code's internal architecture. Brings Claude Code-level development quality and speed to any project through agents, hooks, skills, rules, and CLAUDE.md templates.

Installation

# 1. Add the marketplace
/plugin marketplace add Plusing-net/plusing-claude-skills

# 2. Install the plugin
/plugin install claude-code-patterns@plusing-marketplace

Note: For private repositories, set export GITHUB_TOKEN=$(gh auth token) first.

Local testing (development)

claude --plugin-dir /path/to/claude-code-patterns

Quick Start

1. Run /claude-code-patterns:init-harness to set up CLAUDE.md + .claude/rules/
2. Hooks auto-activate on file edits (format, typecheck, test)
3. Agents auto-delegate on matching tasks (review, debug, security, architecture)

Components

Agents (4)

Specialized AI agents with persistent project memory.

Agent	Model	Purpose
code-reviewer	sonnet	Code quality, reusability, efficiency review
debugger	inherit	Root cause analysis and bug fixing
security-auditor	sonnet	OWASP Top 10 security audit
architect	opus	Architecture analysis and design proposals

Hooks (5 scripts)

Automatic guardrails on file edits. No commands needed.

Hook	Event	What it does
agent-guard.sh	PreToolUse(Agent)	Blocks agents with insufficient context (<100 chars)
lint-guard.sh	PreToolUse(Write|Edit)	Blocks writes to lock files and build artifacts
auto-format.sh	PostToolUse(Write|Edit)	Runs biome/prettier/ruff automatically
typecheck.sh	PostToolUse(Write|Edit)	Runs tsc/pyright (async, blocks on error)
auto-test.sh	PostToolUse(Write|Edit)	Finds and runs related tests (async, blocks on failure)

Skills (5)

Workflow skills invoked manually or by Claude automatically.

Skill	Trigger	Purpose
/deep-think	"think deeply", "analyze tradeoffs"	Structured reasoning with reversibility lens
/design-first	"design first", "plan only"	Design-only mode (no implementation)
/dev-flow	"build this feature", "implement"	Full flow: design → implement → verify → wrap-up
/parallel-attack	"do all at once", "batch"	Multi-agent parallel task execution
/plugin-dev	"create a plugin"	Plugin scaffold generator

Commands (3)

Manual utility commands.

Command	Purpose
/init-harness	Generate CLAUDE.md + .claude/rules/ for the project
/health-check	Diagnose project setup (git, rules, tests, formatters)
/run-hooks-test	Test all hook scripts with dummy input

Rules Templates

Copied to .claude/rules/ by /init-harness.

File	Scope	Applies to
quality.md	All files	Negative instructions, conclusion-first
safety.md	All files	Reversibility lens, destructive op guards
testing.md	**/*.test.* only	Test conventions
api.md	src/api/** only	Security, validation

Architecture

CLAUDE.md (project info) ← always loaded
    ↓
.claude/rules/ (quality rules) ← always loaded, path-scoped
    ↓
Hooks (auto-format, typecheck, test) ← fire on Write|Edit
    ↓
Agents (review, debug, security, design) ← auto-delegate with memory
    ↓
Skills (design-first, dev-flow, parallel) ← manual or auto-invoke

Japanese Version

See claude-code-patterns-ja for the Japanese version.

Author

Tatsuru Okada — CEO at Plusing Inc., an AI technology company.

License

MIT