Install

npm install -g tasuki

Or manually:

git clone https://github.com/ForeroAlexander/Tasuki.git
cd Tasuki && bash install.sh

Requirements: bash, git, python3. Optional: node/npx, jq, curl.

Or via Homebrew (installs Python automatically):

brew tap ForeroAlexander/tasuki
brew install tasuki

The Problem

Without Tasuki, your AI assistant:

• Invents file paths that don't exist in your project
• Skips tests and goes straight to implementation
• Ignores security — introduces SQL injection, hardcoded secrets
• Repeats mistakes — task #89 has the same bug as task #45
• Has no process — you say "write tests first" and it ignores you

With Tasuki, hooks mechanically block those behaviors. No tests? Edit blocked. No plan? Implementation blocked. No agent context loaded? Code changes blocked. It's not a suggestion — it's enforcement.

Quick Start

cd your-project
tasuki                    # scans stack, generates agents, ready

Close and reopen your AI tool so it loads the new hooks. Then:

# In your AI tool chat, prefix with "tasuki" to activate the pipeline:
"tasuki: add user authentication with JWT"

Important: The pipeline activates when you say "tasuki" in your prompt. Without it, your AI works normally. With it, the full pipeline runs — plan first, tests first, security audit, code review. You control when to use it.

The pipeline runs automatically — 9 agents, sequential:

Planner → QA (TDD) → DB Architect → Backend Dev → test checkpoint → Frontend Dev → Security (OWASP) → Reviewer (quality gate) → DevOps → done.

Agents

| Agent | Role | Stage | |-------|------|-------| | Planner | Architecture, PRDs, task decomposition | 1 | | QA | TDD enforcement, test suites | 2 | | DB Architect | Schema design, migrations | 3 | | Backend Dev | APIs, services, business logic | 4 | | Frontend Dev | UI with design preview | 5 | | Debugger | Root cause analysis (reactive) | 5.5 | | Security | OWASP audit, variant analysis | 6 | | Reviewer | Quality gate, 3-round fix loop | 7 | | DevOps | Docker, CI/CD, deploys | 8 |

Each agent is a 250+ line specialist. On Claude Code, agents run as real teammates via Agent Teams — separate context windows, per-agent model selection, shared task list. On other tools, agents run as role-switching within the same context.

Thinking agents (planner, security, reviewer) use the strongest model. Execution agents use fast models. ~60% token savings.

Memory

Two-layer system — not a traditional RAG:

• Layer 1 — Wikilinks: each agent reads only its memories via [[links]]. Zero cost, offline, human-readable.
• Layer 2 — Deep Memory: vector search over schema, APIs, plans, git history. On-demand via MCP. Local SQLite, $0.

Scales from wikilinks-only to pgvector without changing agent logic.

tasuki vault sync          # index project into deep memory
tasuki vault query "auth"  # semantic search across everything

Commands

tasuki                     # onboard or show status
tasuki dashboard           # interactive dashboard (localhost:8686)
tasuki score "task"        # complexity analysis (1-10)
tasuki doctor              # diagnose + auto-fix
tasuki vault stats         # knowledge graph metrics

# Setup
tasuki init <stack> <name>           tasuki onboard [path] [--target=X]
tasuki adapt <target>                tasuki validate
tasuki monorepo                      tasuki ai

# Execution
tasuki mode <fast|standard|serious>  tasuki score "task"
tasuki route "task" [mode]           tasuki cost "task" [mode]
tasuki progress                      tasuki dashboard

# Memory
tasuki vault <stats|search|sync|query>
tasuki facts                         tasuki error "desc" --agent X
tasuki errors [list|clear]           tasuki discover

# Plugins
tasuki plugins                       tasuki install <type> <name>

# Team
tasuki vault push                    tasuki vault pull
tasuki export                        tasuki import <file.tar.gz>
tasuki snapshot <name>               tasuki notify <setup|test>

# Maintenance
tasuki cleanup [--all]               tasuki restore [--all]
tasuki doctor [--fix]                tasuki hooks <install|uninstall>

Stacks

Auto-detected: FastAPI, Django, Flask, Next.js, SvelteKit, Nuxt, Express, NestJS, Rails, Gin, Spring Boot, Laravel, Generic.

Learn More

• Landing page — pipeline visual, memory architecture, before/after demo
• CONTEXT.md — full architecture guide
• CONTRIBUTING.md — how to add profiles, agents, plugins

License

MIT