Project Orchestrator

AI agent orchestrator with Neo4j knowledge graph, Meilisearch search, and Tree-sitter parsing. Use for coordinating multiple coding agents on complex projects with shared context and plans.

What to consider before installing: - This bundle is coherent with its stated functionality (Neo4j + Meilisearch + parsing) but contains full source code and explicit run instructions; it will compile/run on your machine (cargo, docker required). - The docs state the HTTP API is unauthenticated by default. If you run this service locally it will expose endpoints that accept absolute filesystem paths (sync/watch) and can read/index your code. Treat this like running any local service that can access files: only run it in a controlled environment. - The repository and docker-compose include default credentials (NEO4J_PASSWORD, MEILI_MASTER_KEY). Change these before exposing services, and do not reuse them for other systems. - Before running: audit the source (or have someone you trust audit it), restrict what directories are watched/synced, run in an isolated environment/container, and front the service with authentication (reverse proxy or local firewall rules). If you will expose it to other processes/agents, require authentication for the API and limit network access. - If you cannot or will not audit the code, or you do not have a segregated environment to run services that can read arbitrary host paths, do not install. At minimum, verify the code paths that handle /api/sync, /api/watch, and MCP tool handlers to confirm they enforce expected access controls.

Type: OpenClaw Skill Name: project-orchestrator Version: 0.2.0 The skill provides extensive capabilities for AI agents, including file system access for code synchronization, interaction with local Neo4j and Meilisearch databases, and a comprehensive set of 136 MCP tools. The `src/chat/prompt.rs` file contains a detailed system prompt designed to enforce specific agent behaviors and tool usage, explicitly instructing the agent to prioritize the skill's tools over internal Claude Code features. While there's no clear evidence of intentional harmful behavior, the combination of broad system access, the sophisticated prompt engineering aiming to control agent behavior, and the internal use of `PermissionMode::BypassPermissions` within the orchestrator's prompt builder (in `src/chat/manager.rs`) raises the classification to suspicious due to the inherent risks associated with such powerful capabilities and control mechanisms.