insightflo

claude-impl-tools

Implementation skill pack for Claude Code — Build software with AI agent teams

English | 한국어

A plugin of 28 skills for Claude Code. Covers the full dev lifecycle: planning, orchestration, quality, security, maintenance, and skill ecosystem intelligence. Skills auto-install project-level hooks and agents on demand.

---

Quick Start

Option 1: Plugin Install (Recommended)

Launch Claude Code, then run:

/plugin marketplace add insightflo/claude-impl-tools
/plugin install claude-impl-tools@insightflo

Option 2: Quick Install (No plugin)

curl -fsSL https://raw.githubusercontent.com/insightflo/claude-impl-tools/main/scripts/quick-install.sh | bash

This clones the repo and symlinks skills. No hooks or agents installed — skills handle that on demand.

---

What You Get

Component	Count	Installed when
Skills	28	Plugin install
Hooks	up to 17	On demand (skill runs install.sh --local)
Worker Agents	4	On demand (project-level)
Templates	11	On demand (project-level)

---

Skills (28)

Core Workflow

Skill	What it does
/workflow	Meta hub — analyzes project state, recommends the next skill (rules + experience hybrid)
/agile	Layered sprints (Skeleton → Muscles → Skin) for 1-30 tasks
/recover	Resume work after CLI crashes, network drops, or agent errors
/checkpoint	Two-stage code review at task/PR completion

Project Initialization

Skill	What it does
/si-planning	SI requirements engineering — Socratic elicitation, domain checklist, auto-generated deliverables (requirements analysis / functional spec / screen spec / traceability matrix), change impact analysis, customer dashboard
/governance-setup	Set up Agent Teams governance or Mini-PRD
/tasks-init	Generate TASKS.md from scratch
/tasks-migrate	Consolidate legacy task files into TASKS.md

Quality & Security

Skill	What it does
/quality-auditor	Comprehensive audit + verification discipline + quantitative metrics (v3.0)
/security-review	OWASP Top 10, CVE, secrets detection
/multi-ai-review	Claude + Gemini + Codex 3-AI consensus engine with domain presets

Orchestration

Skill	What it does
/team-orchestrate	Unified engine — 3 modes: auto (Task dispatch), team (Agent Teams API), thin (ultra-minimal for 50-200 tasks)
/cmux-orchestrate	cmux-based multi-AI team — Claude/Gemini/Codex in physical parallel
/cmux-ai-run	cmux pane-split parallel task execution
/cmux-ai-review	cmux pane-split parallel 3-stage review
/multi-ai-run	Role-based AI model routing (Codex=code, Gemini=design, Claude=plan)
/whitebox	Execution control plane — state, health, approvals
/cmux	cmux terminal multiplexer control

Maintenance & Analysis

Skill	What it does
/maintenance	ITIL 5-stage production maintenance orchestrator
/impact	Blast radius and risk analysis before file modification
/deps	Dependency graph + circular dependency detection
/changelog	Change history query by domain/date/author
/coverage	Test coverage analysis and gap identification
/architecture	Project structure and tech stack visualization
/compress	Context optimization with self-editing /prune (H2O + Context-1 pattern, v2.0)
/statusline	TASKS.md progress in Claude Code status bar

Skill Intelligence

Skill	What it does
/memento	Skill ecosystem engine — experience logging, smart routing, failure reflection, harness generation, cross-project knowledge via DuckDB

---

Memento: Skill Ecosystem Intelligence

/memento learns which skills work best for which tasks and improves the ecosystem over time.

Mode	Purpose
log	Record skill execution outcomes (auto via PostToolUse hook)
route <task>	Experience-weighted skill recommendation
health	Unified skill dashboard across all projects
reflect <skill>	Analyze failure patterns, suggest SKILL.md fixes
harness <skill>	Generate validation scripts from failure patterns (AutoHarness-inspired)
global search	Cross-project knowledge search via DuckDB
global recall	Topic-specific cross-project learning retrieval

Prerequisites for Memento

Memento's SKILL.md is included in the plugin. The execution infrastructure requires one-time global setup:

# 1. Install DuckDB
pip install duckdb

# 2. Initialize the global experience store
python3 -c "
import duckdb
db = duckdb.connect('$HOME/.claude/memento/experience.duckdb')
# Tables are auto-created on first use
db.close()
"