vanara-orchestratelisted

# vanara-orchestrate — outcomes, not one-shots A single agent answers a question. **Orchestration delivers an outcome**: it chains the right specialists into a pipeline with **gates** between stages, so the work only advances when the previous stage actually succeeded — and nothing risky ships without a sign-off. This is the difference between "here's a suggested fix" and "the bug is fixed, tested, reviewed, committed." You run in the main Claude Code context, so you can invoke installed agents (via the Task tool, or the `vanara-route` skill) one after another and enforce the gates between them. ## The core loop For any goal, run **plan → act → gate → advance**, repeating until done: 1. **Plan the pipeline.** Pick the named workflow that fits the goal (see [references/workflows.md](references/workflows.md)), or compose one from installed agents. State the stages and their gates up front, in one short line each. 2. **Run one stage.** Invoke exactly one agent for the current stage and let it finish. 3. **Check the gate.** Before advancing, verify the stage's **exit condition** holds (tests green, a failing test now written, no blocking review findings, no new secrets — see [references/gates.md](references/gates.md)). If it doesn't hold, do **not** advance: loop back, fix, and re-check. A gate is a hard stop, not a suggestion. 4. **Checkpoint.** Record the stage result so the run is auditable and resumable: ```bash node .claude/skills/vanara-orchestrate