Initiative - Autonomous AI task orchestration for Claude Code
npx claudepluginhub ianmcvann/initiativeAutonomous AI task orchestration for Claude Code
Start it before bed. Wake up to a codebase that's been analyzed by 8 specialist agents, tasks prioritized and executed in parallel worktrees, with commits ready for review.
Initiative is an autonomous task orchestration plugin for Claude Code. It turns Claude into a never-stopping background worker that continuously pulls tasks from a priority queue, executes them in isolated git worktrees, and when the queue runs dry, convenes a panel of 8 expert agents to analyze your project and generate new high-impact tasks automatically. It doesn't just fix problems -- it builds forward.
Initiative runs as an MCP server that provides task management tools to Claude Code. The orchestration loop is simple:
# Clone the repository
git clone https://github.com/ianmcvann/initiative.git
cd initiative
# Add as a Claude Code plugin
claude plugin add ./
# Add the Initiative marketplace
claude plugin marketplace add ianmcvann/initiative
# Install Initiative plugin
claude plugin install initiative@initiative
/initiative:start
Initiative provides seven commands:
/initiative:start [direction]Start autonomous orchestration. Optionally provide a direction to seed the initial task:
/initiative:start
/initiative:start "Add authentication to the API"
/initiative:stopStop the orchestration loop and show a final status summary:
/initiative:stop
/initiative:statusCheck the current state of the task queue:
/initiative:status
Shows total tasks, counts by status, average completion time, throughput, and titles of any in-progress tasks.
/initiative:add [task description]Add a task to the queue without starting the orchestration loop:
/initiative:add "Refactor the database layer to use connection pooling"
/initiative:clearClear all pending tasks from the queue:
/initiative:clear
/initiative:view [task_id]View details of a specific task by ID:
/initiative:view 42
/initiative:historyShow completed and failed task history:
/initiative:history
When the task queue is empty, Initiative doesn't just sit idle -- it convenes a panel of 8 specialist agents that analyze your project in parallel, each from a different perspective. This is what makes Initiative more than a task runner: it's a team of experts that continuously discovers what your project needs next.
| Expert | Focus |
|---|---|
| Security | Vulnerabilities, input validation gaps, exposed secrets, insecure defaults |
| Architecture | Code structure, coupling, separation of concerns, scalability bottlenecks |
| Testing | Coverage gaps, missing edge cases, flaky tests, untested error paths |
| Documentation | Missing or outdated docs, undocumented APIs, incomplete guides |
| Performance | N+1 queries, unnecessary computation, missing caching, memory leaks |
| Product Management | Missing features, incomplete workflows, user value gaps, feature prioritization |
| UX | Confusing interfaces, poor feedback, accessibility, inconsistent patterns |
| Marketing | Value proposition, examples, comparisons with alternatives, distribution |
After all 8 experts report back, Initiative synthesizes their findings: deduplicates overlapping concerns, prioritizes across domains (critical security > product gaps > architecture > UX > testing > docs > performance > marketing), chains related tasks with dependencies, and generates 3-5 concrete, actionable tasks. At least half of generated tasks are additive -- new features and capabilities, not just fixes.
src/initiative/
models.py # Task dataclass and TaskStatus enum
database.py # SQLite task store with priority queue
server.py # FastMCP server exposing task management tools
Task Store -- SQLite database with WAL journaling. Tasks have a title, description, priority (0-1000, higher = more urgent), status, tags, dependency tracking, auto-retry support, and timestamps. The priority queue returns the highest-priority pending task first (ties broken by creation time), skipping any task whose dependencies haven't completed yet.
MCP Server -- Built on FastMCP, exposes 16 tools over stdio transport: