approving
worthwhileVisual multi-agent orchestration with sandboxed execution and human approval gates—valuable infrastructure for teams, overkill for single-agent tasks.
What it is
An open-source orchestrator that lets you design multi-agent workflows as visual finite-state machines, run each agent in an isolated Docker sandbox, pause for human approval at critical nodes, and roll back on failure. Supports Cursor, Claude Code, CodeBuddy, and Trae as backends.
How it differs from vanilla Claude
Claude alone has no built-in multi-agent orchestration, state machine, sandboxing, or human-in-the-loop gates. Approving provides infrastructure for these patterns that a vanilla agent cannot replicate without external tools.
Skill, plugin, or workflow shift?
Standalone app—it's a full web UI + backend + sandbox runtime you deploy and manage, not a plugin or skill you slot into an existing agent.
Devil's advocate — is this just complexity?
For simple one-agent tasks, this is heavy overkill—Docker sandbox, FSM, human gates add cost for no benefit. You can approximate multi-step workflows by chaining Claude calls with careful prompts and manual oversight. The project is early; bugs and missing features may outweigh the structure it provides. Is the state machine really needed, or is it complexity for its own sake? Many teams will find the setup friction too high for the marginal gain.
What would make it better
1) Pre-built template workflows for common patterns (PR review, content generation pipeline) to reduce blank-slate friction. 2) A cloud-hosted version so users skip Docker setup. 3) Integration with version control for workflow definitions (e.g., store graphs as YAML). 4) Better error messages when sandbox images fail to pull. 5) Support for local non-Docker sandboxes for lighter execution.
The honest case for it
If you need safe, auditable multi-agent pipelines where human approval gates prevent costly errors, Approving provides structured infrastructure that ad-hoc scripting cannot match. The visual canvas makes complex workflows comprehensible, and sandboxed execution prevents cross-contamination between agents. For a team running production agent systems, this is the right foundation.
Who it's for
Audience fit
Depth and leverage for a technical engineer who wants to understand it and level up their workflow — not just offload work.
Value for someone who wants a more capable tool without the technical depth — accessible, does-it-for-you.
Engineers building multi-agent pipelines get real leverage from visual state machines, sandboxed execution, and human gates. Vibe coders will find the Docker setup and agent configuration too heavy.