dsh-harvest

Researching with an agent in DSH has never been as simple as throwing a prompt at a search engine. You end up scanning platforms one by one, fighting anti-scraping walls, manually verifying sources, and watching the whole pipeline stall the moment a single channel times out. dsh-harvest packages this entire workflow into a native DSH plugin: five tools chained as Scout, Extract, Verify, and Audit complete multi-platform discovery, article retrieval, cross-source assertion checking, and five-dimension source credibility scoring in a single call, each step emitting structured JSON that the agent can read directly.

Eleven data channels are built in, spanning GitHub, Web search, Twitter, Reddit, Xiaohongshu, YouTube, Bilibili, V2EX, LinkedIn, Telegram, and LINUX DO. When no specific platform is specified, all channels are scanned in parallel. The core design principle is graceful degradation: if any channel fails because a CLI is missing, a virtualenv is absent, or a session has expired, it simply logs [SKIP] and moves on, never blocking the main conversation. The extract step also auto-detects RSS feeds and Xiaoyuzhou podcast URLs, extending coverage to thirteen platforms in total.

Who it is for: DSH workflows that run competitive-intelligence sweeps, technology-stack selection research, or community sentiment monitoring, as well as any operator who wants their agent's research to be auditable, source-traceable, and resilient to partial failures. It does not replace your judgment; it standardizes the dirty work of cast the net, retrieve, verify, and score, so you can focus on the conclusions.