Последовательные многоагентные конвейеры с JSON-стримингом

orchestration

stream-chain is a Claude Code skill that orchestrates sequential multi-agent pipelines by streaming each step's full output as context into the next, enabling sophisticated chained data transformations. It offers two modes: `run` for custom prompt sequences (minimum two prompts required) and `pipeline` for four predefined workflows — analysis, refactor, test, and optimize — each executing three targeted steps. For example, the `test` pipeline covers coverage analysis, test case design, and unit test generation with assertions. All commands run via `claude-flow stream-chain` and support `--verbose`, `--debug`, and `--timeout` flags per step. The skill is built for developers automating security audits, legacy code modernization, technical debt assessment, or structured data processing where output must flow reliably from one agent stage to the next.