Сбор внешней документации и best practices

research

research-agent is a Claude Code skill that gathers external documentation, best practices, and library API references using MCP tools: Nia, Perplexity, and Firecrawl. It is not user-invocable — it is spawned by other agents and receives a research question, context, and a handoff directory as inputs. Library documentation queries go to Nia, best-practices and how-to questions to Perplexity, and specific webpage scraping to Firecrawl; each tool is called via Bash scripts using `uv run python -m runtime.harness`. After gathering data from multiple sources, the agent synthesizes findings into a structured handoff file named `research-NN-<topic>.md`, containing key concepts, code examples, recommendations, pitfalls, and a summary for the next agent — typically plan-agent or implement-agent. If a tool fails, the agent notes the error, continues with remaining sources, and marks the handoff status as "partial" rather than stopping entirely.