What problem does it solve?
This Skill solves the context degradation problem that occurs when complex multi-step coding tasks are executed by AI agents. As context windows fill with full specs, plans, and codebases, output quality drops, tokens are wasted, and failures compound. Orchestra prevents this by decomposing work into a directed acyclic graph of atomic subtasks, each dispatched with a curated prompt containing only the specific objective, relevant spec sections, upstream summaries, and files to touch.
Core Features & Use Cases
- DAG-based Decomposition: Automatically breaks specs, plans, or raw prompts into atomic tasks with explicit dependencies and wave-based parallel execution.
- Context Curation: Assembles minimal prompts for each sub-agent containing only the objective, acceptance criteria, relevant spec slices, upstream results, and necessary files within configurable token budgets.
- Evidence Verification: Runs independent verifier sub-agents to validate completed tasks through screenshots, test output, and build artifacts.
- Cross-Repository Orchestration: Dispatches tasks across multiple sibling repositories with isolated git worktrees per agent.
- Use Case: Imagine implementing a full authentication system. Instead of dumping the entire spec into one agent, Orchestra decomposes it into focused tasks, runs independent ones in parallel, verifies each with evidence, and summarizes results for downstream consumers.
Quick Start
Use the orchestra skill to decompose your specification or prompt into a DAG of tasks and dispatch focused sub-agents to execute them in parallel with minimal context.
