Claude Code Enhancements

A collection of custom agents and skills that extend Claude Code with intelligent automation and autonomous operation capabilities.

What's Inside

Agents

Agent	Description
Agent Prompt Architect	Elite prompt engineer that transforms task requirements into precisely-engineered agent configurations. Handles requirements analysis, expert persona design, system prompt architecture, and validation — outputs production-ready agent specs.
AI Improvement Agent	Evaluates, interrogates, and improves AI agent configurations. Provides structured improvement recommendations with options and reasoning before making any changes. Works with both Claude Code agents and ServiceNow AI configurations.
Deployment Agent	Analyzes pre-built agent configuration files and systematically adapts them to fit your specific environment. Handles cross-platform path adaptation, data source migration, team/stakeholder swaps, and domain customization while preserving proven functional logic.

Skills

Skill	Slash Command	Description
Intelligent Model Router	/smartroute	Autonomous model routing that analyzes task complexity (0-15 scoring), decomposes work into planning (Opus), execution (Sonnet), and simple (Haiku) phases. Achieves 60-80% cost savings vs all-Opus while maintaining quality where it matters.
Fully Autonomous Mode	/fullyautonomous	Status dashboard for autonomous mode. Displays current state, active behavioral directives, and safety boundaries that are always enforced.
Auto-On	/auto-on	Enables fully autonomous mode — bypasses permission prompts for the session. Backs up current permissions, enables all behavioral directives, and maintains a full audit log.
Auto-Off	/auto-off	Disables fully autonomous mode — restores standard permission prompts. Reverts behavioral directives and logs the action.

Installation

Copy the contents into your Claude Code configuration directory:

# Agents
cp -R agents/ ~/.claude/agents/

# Skills
cp -R skills/ ~/.claude/skills/

Restart Claude Code to pick up the new configurations.

Directory Structure

.
├── README.md
├── agents/
│   ├── agent-prompt-architect.md      # AI agent configuration designer
│   ├── ai-improvement-agent.md        # Agent interrogator & improvement advisor
│   └── deployment-agent.md            # Cross-platform agent config adapter
└── skills/
    ├── auto-off/
    │   └── SKILL.md                    # Disable autonomous mode
    ├── auto-on/
    │   └── SKILL.md                    # Enable autonomous mode
    ├── fullyautonomous/
    │   ├── SKILL.md                    # Autonomous mode status & docs
    │   └── autonomous-config.json      # Runtime configuration state
    └── intelligent-model-router/
        └── SKILL.md                    # Smart model selection engine

How It Works

Intelligent Model Router

Routes every task to the optimal model based on complexity scoring:

• Simple (0-3) — File reads, lookups, formatting → Haiku (fastest, cheapest)
• Standard (4-8) — Code generation, analysis, refactoring → Sonnet (balanced)
• Complex (9+) — Architecture, multi-phase workflows → Opus (most capable)

Planning is always Opus. Execution defaults to Sonnet. Simple subtasks use Haiku.

Autonomous Mode

A three-part system for controlling permission behavior:

1. /fullyautonomous — Check current status
2. /auto-on — Enable (bypasses confirmation prompts, maintains audit log)
3. /auto-off — Disable (restores standard permission flow)

Safety boundaries are always enforced regardless of mode — no credentials, SSH keys, API tokens, or destructive operations on protected paths.

Agent Prompt Architect

Transforms task requirements into production-ready agent configurations:

1. Discovery — Asks targeted clarifying questions when requirements are ambiguous, too broad, or missing constraints
2. Design — Structures system prompts with identity, responsibilities, methodologies, quality standards, and edge case handling
3. Optimization — Ensures every instruction is specific, actionable, and builds agent autonomy
4. Validation — Verifies the agent can handle its core task independently with built-in quality control

AI Improvement Agent

An interrogation-first approach to agent improvement:

1. Deep read — Analyzes every line of the target agent configuration
2. Strategic questioning — Probes for pain points, goals, and constraints
3. Structured options — Presents 2-4 improvement paths with trade-offs
4. Approval gate — Never modifies an agent without explicit consent

Deployment Agent

An environment adaptation specialist for agent configuration files:

1. Analysis — Reads the source agent file and categorizes every element (paths, data sources, team names, domain config, functional logic)
2. Discovery — Asks targeted questions in logical groups (environment, data sources, team, domain, workflow, output)
3. Customization — Surgically replaces environment-specific elements while preserving proven functional logic
4. Validation — Verifies all paths are absolute, data access methods match the target setup, and no source-environment artifacts remain

Requirements

• Claude Code CLI installed and configured
• Claude API access (Opus, Sonnet, and/or Haiku models)

License

Private repository. All rights reserved.