project-orchestrator_skill

Overview

This skill is a project orchestrator that coordinates multiple AI coding agents using a shared knowledge graph, fast search, and precise parsing. It combines Neo4j for structural context, Meilisearch for semantic code search, and Tree-sitter for accurate ASTs to keep agents synchronized and focused. Use it to manage multi-project workspaces, structured plans, and decision recording across complex codebases.

How this skill works

The orchestrator syncs a project directory into a Neo4j knowledge graph, indexes code and decisions into Meilisearch, and parses files with Tree-sitter to extract symbols, call graphs, and dependencies. Agents request task context or prompts, consult the graph/search results, perform work, and post back decisions and completion events. The system exposes REST endpoints for projects, plans, tasks, sync/watch, code exploration, and agent webhooks.

When to use it

• Coordinating multiple AI agents on a large or fragmented codebase
• Keeping a shared, queryable history of design decisions and rationales
• Performing semantic code search and impact analysis before changes
• Managing structured plans and task dependencies across teams
• Automating continuous sync of code changes into a knowledge base

Best practices

• Keep each repository or product as a separate project to isolate context
• Run Neo4j and Meilisearch in durable, backed-up environments for production
• Use the file watcher to auto-sync active workspaces and avoid stale context
• Record decisions with clear rationale to improve future agent reasoning
• Limit indexed file types and ignored paths (node_modules, target, .git, etc.) to reduce noise

Example use cases

• Orchestrating multiple code-writing agents to implement a GPU backend with interdependent tasks and recorded decisions
• Performing cross-repo impact analysis before refactoring a core library function
• Providing agents with ready-to-use prompts and rich context for a complex feature ticket
• Searching past design decisions about memory or concurrency to inform implementation choices
• Running automated watchers to keep the knowledge graph current during active development