🧠 Mindkit

Universal Multi-Agent Skill Toolkit for AI Coding Agents

A standalone, project-agnostic collection of agent skills, memory architecture, and orchestration patterns. Works with any AI coding agent (Gemini, Claude, GPT, Cursor, Windsurf, OpenCode, etc.) on any codebase.

---

Quick Start

1. Copy into your agent's skill directory

# For OpenCode
cp -r mindkit/skills/* .opencode/skills/

# For Cursor
cp -r mindkit/skills/* .cursor/skills/

# For any agent that reads markdown skills
cp -r mindkit/skills/* <your-agent-skill-dir>/

2. Copy guardrails into your agent's rules

cp mindkit/GUARDRAILS.md <your-agent-rules-dir>/
cp mindkit/rules/* <your-agent-rules-dir>/

3. Set up memory persistence (optional)

mkdir -p .mindkit/memory .mindkit/sessions

The context-manager skill will auto-persist session context here.

---

What's Inside

mindkit/
├── skills/
│   ├── foundation/     → Token Guardian, Orchestrator, Rule Sanitizer
│   ├── technical/      → Frontend, Backend, CLI, DevOps, QA, Code Review
│   ├── business/       → BA, PM, Tech Lead, AI Lead, UX, CTO/CEO
│   └── infra/          → Context Manager, Memory Architect, Topic Indexer, Skill Registry
├── memory/             → Memory architecture, schemas, templates
├── workflows/          → Pre-built multi-agent workflows
├── rules/              → Consolidated rules & quality gates
├── GUARDRAILS.md       → Universal agent constraints
└── README.md           → You are here

Skills Overview

Foundation (Always Active)

Skill	Purpose
Token Guardian	Monitor & optimize token consumption with guardrails
Agent Orchestrator	Route tasks to the right skills in the right order
Rule Sanitizer	Audit & clean rules, PRDs, and prompts

Technical Agents

Skill	Purpose
Frontend Auditor	UI/component health, styles, routes, build status
Backend Inspector	API wiring, DB layer, error handling, bindings
CLI/Node Reviewer	Commands, npm scripts, dependencies, exports
DevOps Auditor	CI/CD pipelines, deployment, publish configs
QA/Testing Agent	Test coverage, pass/fail, critical path gaps
Code Review Agent	Patterns, TypeScript, security, dead code

Business Agents

Skill	Purpose
Business Analyst	Feature completeness vs spec, gap analysis
PM Agent	User journeys, prioritization, release readiness
Tech Lead	Architecture health, tech debt, scalability
AI Lead	AI integration, prompts, pipelines, model strategy
UX Researcher	Usability, info architecture, accessibility
CTO/CEO Strategy	Vision alignment, GTM readiness, risk register

Infrastructure (Parallel)

Skill	Purpose
Context Manager	Persist & load session context automatically
Memory Architect	Retrieve, prune, link memory entries
Topic Indexer	Extract topics, assign semantic markers
Skill Registry	Discover skills, check compatibility, route

---

Memory System

Mindkit includes a memory architecture that lets agents persist learnings across sessions:

• Session Logs — what happened in each session
• Decision Records — why choices were made (with alternatives)
• Semantic Markers — tagged knowledge (#bug, #architecture, #debt, #insight)

Even weak models can use semantic markers to filter for relevant context before starting work.

See memory/ARCHITECTURE.md for full details.

---

Guardrails

All agents follow strict constraints defined in GUARDRAILS.md:

1. Token budget — every skill has S/M/L cost estimate
2. Output cap — max 100 lines unless overridden
3. Structured output — ## Summary → ## Findings → ## Actions
4. Context-first — read summaries before raw files
5. Idempotent — no side effects, safe to re-run
6. Project-agnostic — no hardcoded paths

---

License

MIT