Token-efficient tier-routed agent orchestrator
40+ skills, 19 agents, Obsidian vault
imlazy classifies every AI coding task by complexity and routes it to the appropriate workflow tier. Each tier loads only the skills it needs, saving tokens on simple tasks while keeping full rigor for complex ones.
Single-file edit, typo fix, version bump, config change, factual question — inline execution with verification gate
Multi-file feature or bugfix with clear requirements — light planning, test-driven development, code review
New subsystem, ambiguous scope, architectural design — brainstorming, full planning with subagents, TDD, code review
Knowledge persistence across sessions — initialize vault, save context, search notes, link knowledge
What makes imlazy different from other agent workflows.
Tasks classified by complexity into Quick, Standard, Heavy, or Vault tiers. Each tier loads only the skills it needs. Simple edits cost approximately 1.5K tokens instead of 14K.
Pre-built skills for planning, test-driven development, code review, brainstorming, debugging, API design, frontend and backend development, security review, WordPress with ACF, coding standards, and more.
Specialized agents for planning, architecture, code review, security review, TDD coaching, build fixing, end-to-end testing, documentation, refactoring, database design, and language-specific reviewers for Python, Go, Rust, Java, C++, Kotlin, and PHP.
Persistent knowledge across sessions using Obsidian Flavored Markdown, structured Bases records, JSON Canvas diagrams, CLI vault interaction, and clean web page extraction via Defuddle.
Every skill file stays within a strict word budget. The router loads only what the tier requires — no wasted context, no unnecessary overhead on simple tasks.
Works with Claude Code, OpenCode, and Codex CLI. Self-contained with zero external plugin dependencies — clone the repo and run the install script.
Install imlazy and start using it in under a minute.
git clone https://github.com/hnikoloski/imlazy.git
cd imlazy
# macOS, Linux, WSL
bash install.sh
# Windows PowerShell
.\install.ps1# Unix / WSL
bash tests/run.sh
# Any platform with Python 3
python tests/run-tests.py/imlazy fix the typo on line 5 of README.md
i am lazy, add a /health endpoint that returns {status: ok}
i am lazy, vault init
/imlazy design the database schema for user subscriptionsEvery skill is a focused SKILL.md file with clear instructions and token budgets. Skills are loaded lazily by the router based on task tier.
imlazy-router — Tier classification and dispatchimlazy-bootstrap — Session entry point and trigger detectionimlazy-planning — Implementation plans (light and full modes)imlazy-tdd — Test-driven development with completion gateimlazy-code-review — Code review request and receive workflowimlazy-brainstorm — Requirements gathering and specificationimlazy-debug — Root-cause investigation workflowimlazy-parallel — Cache-friendly parallel agent dispatchimlazy-worktrees — Isolated workspace managementimlazy-learn — Router misclassification logging and adaptationimlazy-vault — Knowledge persistence and retrievalimlazy-wordpress — WordPress and ACF developmentimlazy-security — Security review patternsimlazy-api-design — REST API design patternsimlazy-frontend — Frontend development (React, TypeScript)imlazy-backend — Backend architecture patternsimlazy-standards — Cross-project coding conventionsimlazy-verify — Pre-completion verification workflowimlazy works with every major AI coding assistant harness without modification.
| Harness | Config File | Install Method |
|---|---|---|
| Claude Code (Anthropic) | .claude-plugin/plugin.json | install.sh auto-detects or --target claude |
| OpenCode | opencode.jsonc | Already in-repo — run opencode load |
| Codex CLI (OpenAI) | .codex-plugin/codex.md | install.sh --target codex |
Open source and free forever.
imlazy is released under the MIT License. Use it, modify it, distribute it — no restrictions. See ATTRIBUTIONS.md for third-party acknowledgments.