dsh-knowledge-forge

Agent sessions carry rich engineering memory — what was tried, what users corrected, which path finally worked — but raw transcripts are too long and noisy to stuff into context, while a single lossy summary drops too many details. DSH Knowledge Forge closes that gap: it automatically distills conversation traces into structured, reusable knowledge instead of asking you to cram every history into the prompt window.

The core path is progressive promotion. A first valuable task case becomes a human-readable Wiki page. When a later independent case repeats the same stable workflow, that Wiki is promoted to a native DSH Skill and the Wiki body shrinks to lightweight metadata pointing at the Skill. At recall time the Agent receives only a compact Wiki index and loads full pages or Skill bodies on demand, preserving context space. Everything runs locally: data lives in project-scoped SQLite and Markdown files with zero external service dependency.

Built for developers running long-cycle engineering tasks in DeepSeek Harness, especially teams that repeat similar workflows (deploys, debugging, code review). If you want your Agent to grow smarter about your project over time while keeping knowledge auditable and hand-editable, this plugin offers a lightweight, transparent, local-first solution.