Savi

Michael Knopf's Claude Code Profile

Personal Claude Code plugin with integrated MCP servers, presets, and team-shareable configurations.

Overview

This plugin provides a centralized configuration for Claude Code that can be version-controlled and shared with team members. It includes several powerful MCP servers that extend Claude's capabilities.

Installation

To install this plugin:

# Clone the repository
git clone https://github.com/michaelknopf/claude-code-profile.git

# Link the plugin to Claude Code
# (Follow Claude Code plugin installation instructions)

MCP Configuration Setup

This plugin uses a templating system to keep secrets out of version control. MCP configuration files (.mcp.json) are generated from template files.

Initial Setup

1. Copy the example environment file:

   cp .example.env .env
   

2. Edit .env with your API keys

3. Generate the MCP configuration files:

   python scripts/render_mcp_templates.py
   

Updating Configuration

After modifying template files (.mcp.template.json), regenerate the config files:

python scripts/render_mcp_templates.py

Script Options

• --dry-run - Preview output without writing files
• --allow-missing - Allow missing environment variables (keeps placeholders)

Example:

# Preview what will be generated
python scripts/render_mcp_templates.py --dry-run

# Generate with some missing variables
python scripts/render_mcp_templates.py --allow-missing

Included MCP Servers

Context7

HTTP-based MCP server providing up-to-date documentation and code examples for programming libraries and frameworks.

• Type: HTTP
• URL: https://mcp.context7.com/mcp

Commands

Analysis & Auditing (Read-Only)

These commands analyze your codebase and produce planning documents without making code changes.

/refactor-audit

Scan codebase for design principle violations and generate a prioritized refactoring backlog.

• Output: docs/local/refactor-audit-YYYY-MM-DD-HH-MM.md
• Follow-up: Use /refactor to apply recommended fixes
• Example: /refactor-audit src/

/coverage-audit

Analyze test coverage gaps and create a prioritized test improvement plan.

• Output: docs/local/coverage-audit-YYYY-MM-DD-HH-MM.md
• Example: /coverage-audit coverage.json --top=5

/typing-audit

Audit Python codebase for type safety issues — dicts that should be typed structures, overuse of Any, and other patterns.

• Output: docs/local/typing-audit-YYYY-MM-DD-HH-MM.md
• Example: /typing-audit src/

/review-pr

Review the current branch's PR diff and produce a prioritized audit report of bugs, security issues, design problems, and more.

• Output: docs/local/review-pr-{BRANCH}-YYYY-MM-DD-HH-MM.md
• Example: /review-pr

/smoke

Design or audit smoke tests using established principles. Two modes: design (create new tests) and audit (review existing tests).

• Example: /smoke design src/api/

Fixing & Implementation

These commands make code changes or modify project state.

/fix

Run a command; if it fails, diagnose and make minimal changes until it succeeds (bounded retries). Intelligently routes type errors to the type-fix-planner agent and other errors to the fix-diagnostician agent.

• Example: /fix just test

/ci-fix

Debug failing CI workflows and push fixes until they pass. Automatically follows workflow chains and routes to specialized agents based on error type.

• Example: /ci-fix 123

/refactor

Apply design principle-based refactoring with codebase-wide discovery of similar violations.

• Example: /refactor src/api/handlers.py:42-80 "convert to RequestHandler class"

/py-dep-upgrade

Remove upper bounds from Python dependencies and upgrade to latest versions, fixing any resulting issues.

• Example: /py-dep-upgrade

/pr-reply

Address, reply to, and resolve PR review comments with code changes.

• Example: /pr-reply all

Task Orchestration

Commands for working through structured task backlogs (integrates with beads issue tracker).

/next

Pick up the next ready task and work on it. Use --loop to process up to 10 tasks in batch mode.

• Example: /next or /next --loop

/epic-loop

Loop through all tasks in a beads epic using the Opus→Sonnet pattern: the task-planner agent plans each task, then a Sonnet agent executes it.

• Example: /epic-loop EPIC-1

Documentation

/doc

Generate high-level conceptual documentation for a file or folder, focusing on the "why" rather than the "what."

• Example: /doc src/api/ --output=docs/api-architecture.md

/doc-refresh

Update existing conceptual documentation to reflect the current state of the code it describes.

• Example: /doc-refresh docs/api-architecture.md

/session-doc

Generate a structured document summarizing the current session for future reference.

• Example: /session-doc ~/Documents/sessions

Workflow

/commit

Create contextual git commits. Analyzes your changes and generates an appropriate commit message.

---

Pro tip: Run analysis commands (/refactor-audit, /coverage-audit, /typing-audit) regularly to build improvement backlogs, then execute incrementally with /refactor, /fix, or /next --loop.

Agents

Commands spawn specialized agents via the Task tool for domain-specific diagnosis and planning. Key agents:

Agent	Model	Used By	Purpose
fix-diagnostician	Opus	/fix, /ci-fix	General build/test failure diagnosis
type-fix-planner	Opus	/fix, /ci-fix, /py-dep-upgrade	Type error diagnosis using type safety principles
task-planner	Opus	/epic-loop	Structured implementation planning
test-suite-planner	Opus	/coverage-audit	Test suite design for coverage gaps
refactor-scout	Sonnet	/refactor-audit, /refactor	Codebase-wide pattern/violation search
typing-scout	Sonnet	/typing-audit	Type safety issue discovery

Extending This Plugin

Adding Custom Commands

Create .md files in the plugins/savi/commands/ directory to add custom slash commands.

Adding Agents

Create .md files in the plugins/savi/agents/ directory to add specialized subagents that commands can spawn via the Task tool.

Adding Skills

Create skill directories with SKILL.md files in the plugins/savi/skills/ directory for reusable task implementations.

Adding Hooks

Add hooks.json configuration files in the plugins/savi/hooks/ directory to respond to Claude Code events.

Principle Documents

Add .md files to plugins/savi/docs/ to define principles that agents reference during analysis. Current documents:

• type-safety-principles.md — Used by type-fix-planner and typing-scout
• smoke-test-principles.md — Used by /smoke