AIKIT

Multi-agent template package manager and CLI. Install and publish aikit.toml template packages from GitHub or a local directory, map them into many coding-assistant layouts, scaffold projects with aikit init, and run supported agent CLIs with aikit run. Use aikit check to see which tools and agents are available.

Programmatic gateway: aikit-sdk (Rust) and aikit-py (Python) expose the same agent catalog, path rules, deploy APIs, availability checks, and run/event APIs for your own tools and automation.

Installation

Linux (GNU)

curl -L https://github.com/goaikit/aikit/releases/latest/download/aikit-x86_64-unknown-linux-gnu.tar.gz | tar xz
sudo mv aikit /usr/local/bin/

Linux (MUSL/Universal)

curl -L https://github.com/goaikit/aikit/releases/latest/download/aikit-x86_64-unknown-linux-musl.tar.gz | tar xz
sudo mv aikit /usr/local/bin/

Homebrew (Linux)

brew install goaikit/cli/aikit

Scoop (Windows)

scoop bucket add goaikit https://github.com/goaikit/scoop-bucket
scoop install aikit

Windows

Download from GitHub Releases and add to PATH.

Quick Start

Install Newton template (workspace with scripts)

# Install Newton template from GitHub or local path
aikit install gonewton/newton-templates --ai newton --yes

# This creates .newton/ with:
#   .newton/README.md - Template documentation
#   .newton/scripts/advisor.sh - Planning advisor
#   .newton/scripts/evaluator.sh - Progress evaluator
#   .newton/scripts/post-success.sh - Post-success hook
#   .newton/scripts/post-failure.sh - Post-failure hook

Create a package and deploy to Cursor and Claude

# 1. Create a new package (defines aikit.toml and template layout)
aikit package init my-tools --description "My AI commands"

# 2. Enter the package and add your templates (rules, skills, prompts)
cd my-tools
# Edit aikit.toml and add files under templates/ as needed

# 3. Build the package (produces dist/ or agent-specific zips)
aikit package build

# 4. Publish to GitHub (creates release and uploads assets)
aikit package publish username/my-tools
# Or: push repo first, then aikit release v1.0.0

# 5. Install the package for Cursor (in a project that uses Cursor)
cd /path/to/your-project
aikit install username/my-tools --ai cursor

# 6. Install the same package for Claude (e.g. in another project or --ai claude)
aikit install username/my-tools --ai claude

# 7. Verify: list installed packages and check available agents
aikit list
aikit check

Use an existing project with an AI assistant

# Create a new Spec-Driven Development project with Claude templates
aikit init my-project --ai claude

# Or set up in the current directory for Cursor
aikit init --here --ai cursor

# Install a community package
aikit install username/package-name
aikit list

Commands

Command	Description
aikit init [name]	Initialize a Spec-Driven Development project with AI assistant templates
aikit install <source>	Install packages from GitHub (owner/repo) or local directory (use --ai <agent> to specify agent)
aikit list	Show installed packages (optional: --author, --detailed)
aikit update <pkg>	Update a package to latest version (optional: --breaking)
aikit remove <pkg>	Uninstall a package (optional: --force)
aikit check	Check git, VS Code, and AI agent CLIs availability
aikit run	Run a coding agent with a prompt
aikit version	Show version
aikit package init <name>	Create a new package with aikit.toml
aikit package build	Build distributable package (output: dist/ or .genreleases/)
aikit package publish <owner/repo>	Publish package to GitHub (release and assets)
aikit release <version>	Create GitHub release from .genreleases/ (e.g. v1.0.0)

Creating and publishing packages

# Create package
aikit package init my-tools --description "AI dev tools" --package-version 0.1.0

# Add templates and edit aikit.toml, then build
aikit package build

# Publish (creates GitHub release and uploads)
aikit package publish username/my-tools

# If you use a flow that produces zips in .genreleases/, create the release with:
# aikit release v1.0.0 --notes-file release_notes.md

Configuration

• GitHub auth: Set GITHUB_TOKEN or GH_TOKEN in .env or use --token / --github-token on install, init, or release.
• Package manifest: Each package has an aikit.toml (name, version, description). Required for local installs and for publish.

Example .env:

GITHUB_TOKEN=your_github_token_here

Running agents

The aikit run command allows you to execute AI coding agents directly. It can replace agent CLI wrappers like coder.sh in workflows like Newton.

# Run an agent with a prompt
aikit run --agent opencode -p "Help me refactor this code"

# Read prompt from stdin
echo "Add error handling" | aikit run --agent claude

# Emit structured NDJSON events to stdout (one JSON object per line)
aikit run --agent claude --events -p "Summarize the project"

# Combine --events with --stream for streaming-aware JSON output
aikit run --agent claude --events --stream -p "Refactor this module"

Supported agents: codex, claude, gemini, opencode, agent

Options:

Flag	Short	Description	Default
--agent	-a	Runnable agent key (codex, claude, gemini, opencode, agent)	Required
--model	-m	Model passed to the agent	Omitted unless -m is set; no CLI default (agent binary decides)
--prompt	-p	Prompt to run	Reads from stdin if omitted
--yolo		Auto-confirm, skip checks	false
--stream		Enable agent-native streaming output flags	false
--events		Emit NDJSON event stream to stdout	false

aikit run also accepts the root --debug flag (global on the aikit CLI; it appears in aikit run --help).

Agent selection is CLI-only: -a / --agent is required. aikit run does not read CODING_AGENT or CODING_AGENT_MODEL. When -m is omitted, no model flag is passed to the agent; the spawned agent applies its own defaults or errors if it requires an explicit model.

--stream vs --events:

• --stream: Tunes agent-native partial output flags (e.g. --stream-partial-output, stream-json). Passed through to the agent argv builder.
• --events: Switches the CLI to emit one JSON object per line (NDJSON) to stdout. Process exit code matches the child agent's exit code.
• Both flags can be combined: --events --stream uses events-mode JSON output AND adds stream-partial flags for supported agents.

NDJSON event format (when using --events): each line is one JSON object with agent_key, seq, stream (stdout | stderr), and payload. The payload is one of:

• json_line: parsed JSON object from the agent
• raw_line: UTF-8 text line that is not JSON
• raw_bytes: non-UTF-8 data as a byte array in JSON
• token_usage_line: normalized token counts extracted from a preceding json_line (fields: usage, source, raw_agent_line_seq)

Example lines:

{"agent_key":"claude","seq":1,"stream":"stdout","payload":{"json_line":{"type":"progress","message":"Starting..."}}}
{"agent_key":"claude","seq":2,"stream":"stdout","payload":{"token_usage_line":{"usage":{"input_tokens":100,"output_tokens":50,"total_tokens":150,"cache_read_tokens":null,"cache_creation_tokens":null,"reasoning_tokens":null},"source":"Claude","raw_agent_line_seq":1}}}

On Windows, the Cursor agent is often agent.cmd. If spawn fails, set AIKIT_CURSOR_AGENT to the full path (see aikit-sdk README: Windows Configuration).

Run aikit run --help for the authoritative option reference.

Programmatic use (gateway libraries):

• Rust (aikit-sdk): Add aikit-sdk to Cargo.toml. Use catalog/path/deploy APIs, run_agent for buffered stdout/stderr, or run_agent_events for the same event stream shape as aikit run --events (including optional token_usage_line events). Returns Result<RunResult, RunError>.
• Python (aikit-py): pip install aikit-py. Same surface area from Python: catalog, deploy, run_agent, and run_agent_events_py(..., on_event, ...) for per-event callbacks (schema matches CLI NDJSON).

Supported AI assistants

The catalog covers 18 coding assistants (install/template mapping). Runnable via aikit run are only: codex, claude, gemini, opencode, agent.

CLI-based: Claude, Gemini, Qwen, OpenCode, Codex, Auggie, CodeBuddy, Qoder, Q, Amp, Shai

IDE-based: GitHub Copilot, Cursor, Windsurf, KiloCode, Roo, Bob

Run aikit check to see which are installed on your system (git and VS Code are also checked).

License

Apache License 2.0 - See LICENSE

Need help? Open an issue