Aimi Engineering Plugin
Autonomous task execution for Claude Code with structured JSON task management.
Transform implementation plans into executable user stories, then run them autonomously with full context isolation. Stories with independent dependencies execute in parallel via Team orchestration with git worktrees. Each story gets its own agent with automatic state tracking.
Table of Contents
Installation
Claude Code
No external dependencies. This plugin is fully standalone.
# Add marketplace and install
claude /plugin marketplace add https://github.com/aimi-so/aimi-engineering-plugin
claude /plugin install aimi-engineering
Verify installation:
claude /plugin list
/aimi:status
OpenCode
Install using the built-in installer script. No external dependencies required.
git clone https://github.com/aimi-so/aimi-engineering-plugin
cd aimi-engineering-plugin
./install.sh --to opencode
This installs the plugin into OpenCode's global config directory (~/.config/opencode/) with full compatibility translation:
- Commands are translated and installed as nested directories (e.g.,
commands/aimi/plan.md) so they appear as /aimi:plan, /aimi:execute, etc.
- Command bodies are fully rewritten for OpenCode compatibility — agent invocations use the Task tool, CLI path references use
OPENCODE_CONFIG_DIR, and error messages point to the OpenCode installer
- Skills are copied to
~/.config/opencode/skills/ with SKILL.md and reference files preserved
- Agents are translated to
~/.config/opencode/agents/aimi-*.md with model fields preserved
- Permissions are auto-configured in
opencode.json for autonomous Bash execution
- Plugin source (CLI scripts, hooks) is copied to
~/.config/opencode/plugins/aimi-engineering/
- context7 MCP server is added to
opencode.json
AIMI_PLUGIN_DIR is set in your shell profile
After installation, restart your shell or run:
export AIMI_PLUGIN_DIR="$HOME/.config/opencode/plugins/aimi-engineering"
Known Limitations
disable-model-invocation is not supported in OpenCode. The installer prepends a side-effect warning to command bodies as a workaround, instructing the model not to invoke sub-agents autonomously.AskUserQuestion tool is not available in OpenCode. Commands that use it are rewritten to use natural conversation prompts instead (asking the user directly in the response).- Custom
subagent_type (e.g., subagent_type: researcher) is not yet supported in OpenCode. Agents are installed as general-purpose with the agent prompt inlined into the Task tool invocation.
Project-Level Install
To install into .opencode/ in your project directory instead of globally:
./install.sh --to opencode --project
Uninstall
./install.sh --uninstall --from opencode
Preview Changes
./install.sh --to opencode --dry-run
Environment Variable
The AIMI_PLUGIN_DIR environment variable points to the installed plugin directory so that commands can locate CLI scripts, skill references, and agent definitions. The installer sets this automatically. Do not set this variable manually unless you have a custom installation layout.
Quick Start
# 1. Brainstorm your feature
/aimi:brainstorm Add user authentication with email/password
# 2. Generate tasks directly
/aimi:plan Add user authentication
# 3. Execute all stories autonomously
/aimi:execute
# 4. Review the implementation
/aimi:review
Commands
| Command | Description | Usage |
|---|
/aimi:brainstorm | Explore ideas through guided brainstorming | /aimi:brainstorm [feature] |
/aimi:plan | Generate tasks.json directly from feature description | /aimi:plan [feature] |
/aimi:deepen | Enrich tasks.json stories with research insights | /aimi:deepen [tasks-path] |
/aimi:status | Show current task execution progress | /aimi:status |
/aimi:next | Execute the next pending story | /aimi:next |
/aimi:execute | Run all stories autonomously (parallel for v3, sequential for v2.2) | /aimi:execute |
/aimi:review | Multi-agent code review with findings synthesis | /aimi:review [PR or branch] |
/aimi:swarm | Execute multiple tasks.json files in parallel Docker sandboxes | /aimi:swarm [--file path] [--max N] |
Command Details
/aimi:brainstorm
Standalone brainstorm workflow with codebase research and Ralph-style batched multiple-choice questions. Explores requirements and approaches interactively before committing to implementation.
aimi-so