dsh-ima-copilot

Tencent IMA is a well-regarded knowledge-base product, yet its official public-knowledge skill only performs keyword search by file title, leaving semantic retrieval largely unaddressed. Agents running inside DeepSeek Harness therefore often receive irrelevant or empty results, forcing users to manually browse files and guess keywords. dsh-ima-copilot wraps IMA knowledge-base Q&A into a native DSH tool called ima_ask: the agent passes a natural-language question and gets back a semantically matched answer body together with citation details such as source ID, title, and knowledge-base name, closing the biggest gap in the official skill.
The plugin ships as a self-contained DSH bundle with no MCP server, FastMCP, Python runtime, or persistent subprocess to manage. Authentication cookie, business key, and knowledge-base IDs are all configured through a built-in Web settings page, and changes take effect immediately without restarting DSH. Runtime safeguards include a 300-second request timeout, three retries on transient failures, a concurrency cap, and proper cancellation handling. Credentials are read and written exclusively via the DSH Credentials API and are never echoed back to the UI.
Best for developers building agents in DeepSeek Harness who need semantic Q&A with citation provenance against Tencent IMA knowledge bases. One caveat: the plugin calls the IMA Web API rather than a public stable interface, so upstream schema, auth, or SSE format changes may require a plugin update.