DeepSeek Harness Plugins · Memory
Plugins that extend DeepSeek Harness (DSH) with vision, workflows, memory, web tools and more.
Indexes 8 pinned Clean Architecture reference repos with FTS5 full-text search, machine-checkable rule checklists, and an audit ledger for one-click baseline retrieval.
A local-first literature workbench for psychology and social science, integrating multi-source discovery, PDF close-reading, systematic review, and evidence export within DeepSeek Harness.
Adds a Skill Manager panel to the DSH Web UI sidebar for viewing, searching, and deleting local skills under ~/.dsh/skills.
Generates a brief recap card in the background when you switch away from a DSH Web session, summarizing current task, progress, and next steps.
A DSH plugin that leverages Milvus for semantic code search, hybrid retrieval, incremental indexing, call chain analysis, and ADR decision memory for coding agents.
A DSH plugin lifecycle control plane with transaction-based install/uninstall, 7-type conflict detection, layered validation, and a discovery marketplace.
Persistent project memory and architecture analysis for DSH, preserving decisions, TODOs, and context across sessions with local BM25 retrieval.
Automatically distills conversations into knowledge cards with cross-session recall, shared across multiple agents, zero third-party dependencies, SQLite persistence, and human-in-the-loop review.
DeepSeek Harness MySQL session-persistence backend that appends event-sourced session logs transactionally into MySQL, with read/write split, crash recovery, and lazy materialization.
Reproduces agentic IDE Knowledge Center as a dsh plugin, persisting RepoWiki, Knowledge Cards, and Memory into a local Obsidian vault with hybrid search, activity-driven refresh, and AI-based enrichment.
Folds pasted code with path:line headers into compact reference chips in the DSH input box, inlining the full code snapshot to the model on send.
Give your coding agent effectively infinite context: fold completed tasks into short summaries, cut long-session costs, and recall the full original content on demand.