A DeepSeek Harness plugin layer that gives LLM agents a persistent identity: PersonaBots — bots with a persona and memory that outlive any session, chat, or workspace.
Language
PersonaBots
PersonaBot: A first-class bot entity owned by the Host: a persona plus memory that spans sessions, chats, and workspaces, able to hold several sessions at once. Avoid: bot (bare), agent, assistant, robot
Bot as a Person: The principle that a PersonaBot’s continuity comes from its memory files, not from any session history. Avoid: session-scoped identity
Bot slug: A PersonaBot’s filesystem-safe identifier — kebab-case, unique per Host. Avoid: display name, title, handle
Display name:
The human-facing name of a PersonaBot (displayName), distinct from its slug.
Avoid: alias, nickname, username
Persona: The role definition — character, voice, and standing instructions — that shapes how a PersonaBot replies. Human-owned: the Agent may not rewrite it. Avoid: system prompt, character sheet, profile
Bot state:
The activity model in two levels: a session carries the detail (thinking, working, waiting, blocked, done), and the PersonaBot aggregates it (blocked > waiting > working > thinking > idle; done is a session event).
Avoid: status, mood, presence
Avatar: A PersonaBot’s visual representation; the MVP uses deterministic blobatars, Live2D is a later renderer. Avoid: profile picture, skin
Support and execution
Harness: The platform layer — BotHarness — that owns PersonaBots, their memory, state, and workspaces, and exposes them to other plugins. Avoid: framework, runtime, kernel
Host: The single DSH process that runs the plugin and every PersonaBot. Avoid: server, instance, node, worker
Agent: The DSH executor inside one Session. Never a PersonaBot. Avoid: using this word for PersonaBot
Session: One run of work or conversation for a PersonaBot — DSH’s execution unit, with its own progress and working directory. Avoid: conversation, context window, thread
Workspace: A single host directory a Session works in; it maps one-to-one to a DSH workspace. Several workspaces may be grouped in the UI, but a workspace never spans directories. Avoid: project, multi-root folder, group
Delegation: Handing a PersonaBot work from a chat or the roster; the work runs in a Session. Avoid: assignment, task, job
Channel binding: A PersonaBot’s connection to an external surface — an IM app, the sidebar, or a renderer. Avoid: integration, connector
Memory
Memory: A PersonaBot’s persistent knowledge: human-readable Markdown files in a user-configurable memory directory, shared across every session, chat, and workspace. Avoid: knowledge base, vector store, RAG, database, context
MEMORY.md: The entry-point file of a PersonaBot’s memory: persona summary, usage notes, and an index of topic files with one-line summaries. Avoid: index, README, manifest
Topic file: A memory file devoted to one subject — a customer, a process, a decision — under the PersonaBot’s memory directory. Avoid: note, document, page, record
Customer profile: The north-star topic file: one per customer, holding timeline, key facts, commitments, and links to related Attachments. Avoid: CRM record, account, contact sheet
Memory tree injection: The turn-start injection of the memory directory tree — paths, one-line summaries, and updated-at times — with file bodies fetched only on demand. Avoid: memory dump, prefetch, embedding
Memory tool: A model-callable tool that reads, searches, or writes Memory; every Memory change goes through one. Avoid: memory plugin, hook, background job
Tool-write: The rule that Memory changes only when the model explicitly calls a Memory tool; there is no background distillation. Avoid: auto-summary, auto-extract, distillation
Visibility:
Whether a memory entry may surface outside the chat that produced it: shared (any chat) or private (its author’s DMs only).
Avoid: scope, ACL, secret
Attachment: A file or image uploaded into a Chat, archived into the PersonaBot’s workspace at a stable path and referenced from context and Memory. Avoid: upload, media, blob
Chats and replies
Chat:
A Feishu/Lark conversation — group or p2p — that a PersonaBot takes part in, identified by chat_id.
Avoid: room, channel, group (when p2p is meant too)
DM: A p2p Chat between a PersonaBot and one user. Avoid: private chat, PM
Thread: A sub-conversation opened by replying to a message inside a Chat. Avoid: topic, sub-chat, channel
Reply scope: Where a PersonaBot’s answer lands: under the triggering message in the Chat (root), or inside a new Thread. Avoid: reply mode, answer position, visibility
Host and setup
PersonaBot registry: The Host’s record of PersonaBot definitions and their channel bindings. Avoid: config file, database, fleet
Roster: The in-harness panel listing PersonaBots, their state, and their sessions. Avoid: dashboard, bot list
Settings UI: The in-harness DSH settings surface for the setup wizard, plugin settings, PersonaBot management, memory editing, and diagnostics. Avoid: admin panel, dashboard, web console
Access policy: The per-PersonaBot list of users and chats allowed to reach it. Avoid: whitelist, permissions, ACL
Credential reference: A pointer to a Feishu App Secret held by the DSH credentials service; the secret itself never reaches config, repo, or logs. Avoid: secret, API key, token