🤖 Strikethroo

Strikethroo: extensible AI-powered plan and task management with customizable workflows and structured development processes.

Transform complex AI prompts into organized, executable workflows through customizable hooks, templates, and progressive refinement. Works seamlessly within your existing AI subscriptions for Claude Code, Gemini CLI, and Open Code.

🚀 Quick Start

This project ships in two parts: the skills (installed by vercel-labs/skills) and the workspace (initialized by this CLI). Run both:

# 1. Install the Strikethroo skills for your assistant
npx skills add e0ipso/strikethroo

# 2. Initialize the .ai/strikethroo/ workspace
npx strikethroo init --harnesses claude --destination-directory .

The skills give your assistant the planning, decomposition, and execution workflow; the CLI bootstraps .ai/strikethroo/ with hooks, templates, and the hash-tracked diff-on-conflict UX. Each step is independently re-runnable. See the migration guide for upgrade flows from 1.x.

The CLI's init emits the shared workspace plus harness-specific agents (e.g., .claude/agents/ for Claude). Other harnesses rely entirely on the installed skills.

🌐 Serve the Workspace

The web app is Strikethroo's visualization layer — the way to see your plans, tasks, the dependency graph, and the strikethrough done-state instead of reading a tree of Markdown files by hand. It is the other half of the story: the assistant authors plan and task files inside your harness (Claude Code, Gemini CLI, opencode, the desktop app), and the web app is the read-only window onto those same files.

Run a local web app over an initialized .ai/strikethroo/ workspace:

npx strikethroo serve

See The Web App for a full tour of the Board, Plan Detail, dependency graph, archive, and customization screens.

serve boots a dependency-light Node server that hosts the prebuilt single-page app as static files, exposes a read-only JSON API over the workspace model, and streams a coalesced changed event over Server-Sent Events whenever the workspace mutates on disk. Run it from inside an initialized workspace; if none is found it prints guidance to run init and exits without binding.

The viewer is read-only except for one permitted mutation: the archive action. A plan whose tasks are all complete (derived state done) shows an Archive control; confirming it issues POST /api/plans/:id/archive, which moves that plan's directory from plans/ to archive/. It is strictly a directory move — no files are deleted or edited, and only done plans are accepted. This is the manual escape hatch for plans that are done but not yet archived; it does not replace the automatic archival the st-execute-blueprint skill performs on successful completion.

Flags:

• --port <n> — port to bind (default 4317).
• --no-open — do not open the browser on start.
• --workspace <path> — override workspace root discovery.

✨ Key Benefits

• 🔧 Fully Customizable: Tailor hooks, templates, and workflows to your project's specific needs
• 🎯 Extensible Architecture: Add custom validation gates, quality checks, and workflow patterns
• 📋 Structured Workflows: Three-phase progressive refinement with validation gates
• 🔄 Plan Mode Integration: Enhance existing AI assistant features with structured task management
• 💰 Works Within Subscriptions: No additional API keys or costs required

📖 Documentation

🌐 Complete Documentation →

Comprehensive guides covering:

• Installation and configuration
• Customization with hooks and templates
• Workflow patterns and best practices
• Architecture and design principles

🔄 Workflow Preview

Once the skills are installed, invoke the workflow by intent — the assistant auto-loads the matching skill. There are no slash commands to memorize.

Automated end-to-end run:

"Run the full workflow to create a user authentication system."

The st-full-workflow skill handles plan creation, task generation, and blueprint execution in a single pass.

Step-by-step (for manual review between phases):

1. 📝 Create a plan → "Create a plan for a user authentication system" (st-create-plan skill)
2. 🔍 Refine the plan → "Refine plan 1" (st-refine-plan skill — useful when a second assistant should red-team the plan)
3. 📋 Generate tasks → "Generate tasks for plan 1" (st-generate-tasks skill)
4. 🚀 Execute blueprint → "Execute the blueprint for plan 1" (st-execute-blueprint skill)

Progress is driven by the Agent Skills workflow: the st-execute-blueprint skill runs the phases, archives the plan automatically on completion, and you can inspect plan and task files directly under .ai/strikethroo/plans/ (completed plans move to .ai/strikethroo/archive/).

🤖 Supported Harnesses

| Harness | Interface | Setup Time | |-----------|-----------|------------| | 🎭 Claude | claude.ai/code | < 30 seconds | | 🖱️ Cursor | Cursor IDE | < 30 seconds | | 💎 Gemini | Gemini CLI | < 30 seconds | | 📝 Open Code | Open source | < 30 seconds | | 🔮 Codex | Codex CLI | < 30 seconds | | 🐙 GitHub Copilot | VS Code / JetBrains IDEs | < 30 seconds |

📄 License

MIT License - Open source and free to use.