Memory and knowledge · DeepSeek Harness plugins — Page 26

中文 DSH 本地知识复习插件：零配置资料库、证据优先问答、Markdown 摘要与二次分类。

Evidence-backed, human-approved project memory for DeepSeek Harness

Project-scoped AI memory: records pitfalls and user habits per project, injects them into future conversations

Local project, requirement-group, and conversation workbench for DeepSeek Harness Web and Desktop

PRO-LONG-style programmatic memory for DeepSeek Harness: appends every session event to a per-workspace log.txt that the agent retrieves with grep/python. Built-in write probe, /prolong status command and permission-deny counter.

Personal prompt profile engine for DeepSeek Harness: message-level prompt rewriting and automatic profile learning with a bilingual settings panel and conversation dock.

Persistent input memory for DeepSeek Harness: arrow-key history switching in the composer (shell-style) plus automatic draft snapshots that survive crashes or failed submissions.

Local knowledge base for DeepSeek Harness: document CRUD, keyword search and a kb_search model tool.

Conversation history recall for DeepSeek Harness: search the original text of every session (like searching chat records) — literal/fuzzy/semantic retrieval, fully local & offline. AI never forgets what you told it.

DeepSeek Harness plugin: recall sent-but-unread text messages before the model reads them.

Cross-session long-term memory for the DeepSeek Harness: a two-phase extraction/consolidation pipeline over a file-based memory workspace, recall injection, a citation feedback loop, lease-based job semantics, plus a settings page, a sidebar status panel and a memory viewer.

Baigong research continuity tools for DeepSeek Harness: persistent tasks, evidence and stage handoffs across executors.

AutoResearch suite for DeepSeek Harness: arxiv recall, LQS scoring, research state panel, stall guardian, five-persona peer review, plus the research-ui visualization feature-switch control plane.

Runtime metrics and correlation API for DeepSeek Harness observability.

A Git-native research graph for literature-grounded research expansion inside DeepSeek Harness.

Sensitive-information redaction for DeepSeek Harness: masks API keys, bearer tokens, JWTs, private keys and configured secrets in every tool result shown to the model, plus a redact_text tool and a redact RPC. · DSH 敏感信息脱敏插件：自动掩码工具输出中的 API key / token / 私钥 / 配置密钥。

DSH cross-session self-improvement: memory, error/session retrospectives, SOP extraction and improvement proposals

Session-scoped full-duplex voice assistant with tool-controlled drafting and Agent submission for DeepSeek Harness.

DSH plugin: same-workspace cross-session read-only browsing for models — list_sessions / read_session / search_sessions tools with strict cwd scoping, UNTRUSTED snapshot marking, and bounded budgets (optional SQLite FTS full-text search).

Cross-tool session migration plugin for the DeepSeek Harness (dsh): high-fidelity ingest, continue, knowledge extraction, bidirectional export, privacy redaction.

Session reference plugin for DeepSeek Harness: one-click copy of the current session's @session-id reference in the header, an @ mention menu that lists sessions by title, and a session_read tool that reads any local session by @session-id / dsh-session:URI / @title and projects it to readable conversation text — no more download-transfer-unzip to share context between sessions.

Recover deleted/corrupted dsh sessions (session.jsonl.zstd) and memory (memory.db) straight from the raw disk. Battle-tested recovery manual + scripts + a /session-repair command plugin for the dsh web UI.

DSH web plugin: cross-workspace session status alerts — floating capsule toasts (aligned with the composer) above the current workspace when any session completes or enters a pending interaction (approval / ask_user_question / plan-review). Inline approval & question answering right in the toast, click-to-switch-workspace, configurable layout (stack / inline / grouped manager), subagent completion filter, settings page, i18n zh/en.

DSH skill_manage: agent-managed procedural memory. Create/patch/edit/disable/delete skills in ~/.dsh/skills (user scope) or <projectRoot>/.dsh/skills (project scope), with validation, delete guards (agent-created marker, pin, path confinement), disable/enable via disable-model-invocation, and an English trigger-discipline system prompt section. Built on top of skill-filesystem's hot-reload watcher.