Getting Started

Get up and running with Strikethroo quickly. This page covers prerequisites, installation, directory structure, your first workflow, and verification.

Strikethroo transforms complex development requests into organized, executable workflows through a three-step workflow:

  1. Create Plan – Define objectives and clarify requirements
  2. Generate Tasks – Break the plan into atomic tasks with dependencies
  3. Execute Blueprint – Implement tasks with review gates and quality gates

The workflow is delivered through harness-agnostic Agent Skills installed via npx skills add e0ipso/strikethroo. Skills load automatically when their description matches your intent, so you drive the workflow by asking your assistant to plan, decompose, or execute – no commands to memorize.

Prerequisites

  • Node.js: Version 14.0 or higher
  • npm: Comes with Node.js
  • AI Harness: A harness that supports the Agent Skills format (Claude, Gemini, OpenCode, Codex, GitHub, Cursor, etc.)

Installation

No global installation required. Use npx to run both steps directly.

Step 1: Bootstrap the Workspace 

npx strikethroo init --harnesses claude

This creates the shared .ai/strikethroo/ directory (plans, archive, config) and copies the Claude agents into .claude/agents/.

Step 2: Install the Workflow Skills 

npx skills add e0ipso/strikethroo

This installs the Agent Skills that implement the three-step workflow. Skills are harness-agnostic – a single skill works for every harness that supports the Agent Skills format.

To pin a specific release:

npx skills add e0ipso/strikethroo@<tag>

The --harnesses Flag

The --harnesses flag is required when running init. It controls which per-harness artifacts the CLI copies in addition to the shared workspace.

Each harness can have its own agent files deployed to the appropriate location (e.g., .claude/agents/ for Claude). Harnesses without harness-specific agents get the shared workspace only and rely entirely on the installed skills:

# Claude (copies .ai/strikethroo/ + .claude/agents/)
npx strikethroo init --harnesses claude

# Other harnesses (copies .ai/strikethroo/ only; rely on skills)
npx strikethroo init --harnesses gemini

You can pass multiple harnesses in a single run:

npx strikethroo init --harnesses claude,gemini,opencode,codex,github,cursor

All harnesses share the same task management structure (plans, tasks, configurations).

Custom Destination Directory

By default, init writes into the current working directory. Use --destination-directory to target an alternative location:

npx strikethroo init \
  --harnesses claude \
  --destination-directory /path/to/project

Directory Structure

After running both commands, the project layout looks like this:

project-root/
├── .ai/
│   └── strikethroo/               # Shared configuration files
│       ├── plans/                 # Active plans (empty initially)
│       ├── archive/               # Completed plans (empty initially)
│       ├── config/
│       │   ├── STRIKETHROO.md     # Project context (customize this!)
│       │   ├── hooks/             # Lifecycle hooks
│       │   │   ├── PRE_PLAN.md
│       │   │   ├── PRE_PHASE.md
│       │   │   ├── POST_PHASE.md
│       │   │   ├── POST_PLAN.md
│       │   │   ├── POST_TASK_GENERATION_ALL.md
│       │   │   ├── PRE_TASK_ASSIGNMENT.md
│       │   │   ├── PRE_TASK_EXECUTION.md
│       │   │   ├── POST_ERROR_DETECTION.md
│       │   │   └── POST_EXECUTION.md
│       │   ├── templates/         # Customizable templates
│       │   │   ├── PLAN_TEMPLATE.md
│       │   │   ├── TASK_TEMPLATE.md
│       │   │   ├── BLUEPRINT_TEMPLATE.md
│       │   │   └── EXECUTION_SUMMARY_TEMPLATE.md
│       └── .init-metadata.json    # File conflict detection tracking
└── .claude/                       # Claude agents (if --harnesses claude)
    └── agents/
        └── plan-creator.md

The installed skills live wherever your assistant manages them; they are not copied into the project tree.

Your First Workflow

Once installation is complete, you drive the three-step workflow entirely through your assistant. Here is the typical sequence:

  1. Create a plan – Ask your assistant to plan a feature or change. The st-create-plan skill loads automatically, gathers requirements, and produces a work order in .ai/strikethroo/plans/.

  2. Generate tasks – Ask your assistant to decompose the plan into tasks. The st-generate-tasks skill breaks the plan into atomic tasks with dependency mapping and writes them into the plan’s tasks/ subdirectory.

  3. Execute the blueprint – Ask your assistant to execute the plan. The st-execute-blueprint skill works through each phase, running tasks in dependency order with quality gates between phases.

You can also run all three steps in a single uninterrupted sequence by asking your assistant to handle the full workflow end-to-end (the st-full-workflow skill).

For a detailed walkthrough of the day-to-day development cycle, see the Workflow Guide.

Verification

Verify successful installation:

1. Check Directory Structure 

ls -la .ai/strikethroo/

You should see: plans/, archive/, config/

2. Check Claude Agents (if you ran --harnesses claude) 

ls -la .claude/agents/

3. Inspect Plans on Disk

Plans live under .ai/strikethroo/plans/ and completed plans are moved to .ai/strikethroo/archive/ automatically by the st-execute-blueprint skill. Until you create your first plan, plans/ is empty:

ls -la .ai/strikethroo/plans/

4. Confirm Skills Are Installed

Open your assistant and ask it to create a Strikethroo plan. The st-create-plan skill should load automatically based on intent.

Updating

Re-running init

Re-run the init command to update the workspace files to the latest version:

npx strikethroo init --harnesses claude

File Conflict Detection automatically:

  • Compares current files to original versions using SHA-256 hashes
  • Prompts if you have customized files (shows unified diff)
  • Updates unchanged files automatically
  • Preserves your customizations

Force Mode

Bypass conflict detection prompts (useful for automation):

npx strikethroo init --harnesses claude --force

Warning: Force mode overwrites ALL files, including your customizations. Back up custom hooks and templates first.

Updating Skills

Re-run npx skills add e0ipso/strikethroo to pull the latest skills. The two channels (CLI init and skills installer) are independently re-runnable.

Customizing for Your Project

After installation, customize these files for your specific needs:

Essential Customizations

  1. .ai/strikethroo/config/STRIKETHROO.md
    • Add project context (tech stack, coding standards, architecture decisions)
    • Include links to design docs, API specs, or style guides
    • Document project-specific conventions
  2. .ai/strikethroo/config/hooks/POST_PHASE.md
    • Add your quality gates (linting, tests, coverage thresholds)
    • Include deployment steps (staging, production)
    • Add notification steps (Slack, email, dashboard updates)

Advanced Customizations

See the Customization Guide for:

  • Template modifications
  • Hook customization examples
  • Real-world scenarios (React projects, API projects, monorepos)

Next Steps

This site uses Just the Docs, a documentation theme for Jekyll.