Skip to content

Todo Workbench

The Todo feature belongs to ChatSite. ChatTodo supplies reusable task-forest operations and persistence, not another Web host.

Features

  • Left-hand canvas navigation: create, name and switch independent overview/project/idea canvases.
  • Native SimpleMindMap editing: select, double-click/F2 for titles, Tab for children, Enter for siblings, drag/reparent, multiselect, collapse and layout selection. Touch devices support two-finger zoom and pan.
  • Storage and integration documents both SQLite databases, schemas, revisions, idempotency, Python/HTTP contracts and backup boundaries.
  • Nodes are plain text. Optional Markdown/PRD belongs in a separate detail panel; rich-text or arbitrary HTML map nodes are not enabled.
  • Ant Design X conversation panel with user/assistant messages, Markdown/code, copy and change receipts; scope it to a selected branch or the whole board and collapse it on narrow screens.
  • Single-root canvases open directly. Existing forests expose an explicit root selector: editing one branch preserves all other roots. Empty canvases can add a new center node.
  • Private SQLite persistence, atomic batches, revisions, idempotency, history and undo.
  • Structured model edits validated against schema and scope; important changes require confirmation.
  • No model keys in the browser and no arbitrary execution of model-generated code.

Configuration and execution

Install "ChatSite[todo]>=0.1.5,<0.2.0" with pip. The extra resolves the compatible ChatTodo domain dependency (>=0.1.0,<0.2.0) and ChatLogin Web login core (>=0.1.2,<0.2.0). The Python package includes the frontend build; deployment requires neither Node nor a runtime CDN.

Models may use Responses or Chat Completions. Ark Agent Plan supports doubao-seed-evolving through Responses; retain the provider's Plan endpoint without automatic pay-as-you-go fallback. Run configuration checks in the same environment that has Todo installed so ChatEnv can discover its provider.

The ChatEnv provider alias is chatsite-todo; its single storage namespace is ChatSiteTodo. Login credentials, model protocol/base/model/key and runtime paths belong to this configuration boundary. Explicit named profiles neither activate globally nor borrow another account's environment values.

chatsite todo --help
chatsite todo check
chatenv test -t chatsite-todo -I
chatsite todo serve

Checks send one bounded real no-edit model request. The service binds to loopback by default and is exposed through a reverse proxy.

Data and protocol

Browser requests use same-origin session cookies and write-time CSRF validation. /login uses ChatLogin's shared login page assets; /api/login remains compatible with {email,password} and also accepts the shared form's {username,password}. Model requests originate on the server. Stateless Responses calls use per-board local history; unfinished function-call response IDs are not reused as provider-side conversation chains.

Before submitting, the shared login page reads public GET /login/session: anonymous, invalid/expired or logged-out sessions receive 200 {"authenticated":false}. Valid sessions expose only authenticated, email and csrf_token, never passwords, model keys or the raw session token; responses are not cached. Non-authentication failures such as storage errors remain errors, not anonymous responses. The workbench recovery endpoint GET /api/session still returns 401 when unauthenticated. Login and bootstrap URLs honor root_path; a missing or unsafe next defaults to the current mount root in both the page and login response (/ at the site root, for example /mounted/ when mounted). Explicit safe same-site next paths are preserved.

Ordinary conversation may return natural language, Markdown or JSON examples without modifying the board. Edits require a real todo_update tool call with validated arguments; malformed, unknown or multiple tool calls never become executable text. Definite model-generation failures allow a new user request, while missing write receipts retain the original request identity for retry.

Nodes contain id, parent_id, title, status, body, and order. Operations are create, update, move, and delete. Model output cannot select an owner or another board. Request sizes, node counts, content sizes and model responses are bounded.

TodoSettings.data_dir describes the runtime root, normally chatsite/todo under ChatArch home. Directories and SQLite files are private. ChatLogin sessions live in a separate auth.sqlite3; tasks, conversations, proposals and view state stay in the existing business databases. Upgrading to 0.1.5 does not migrate old Todo local sessions, so users sign in once again; business data is not migrated or deleted. Exports do not include login or model credentials.

Development and acceptance

Undo uses server history for both manual edits and applied model changes. Redo uses revision-guarded snapshots from the current page session; other semantic edits or a reload invalidate them. The model endpoint returns completed results, not fake streaming or a nonfunctional cancel action. Native text edits are committed before sending, navigation or logout; ambiguous writes retain the original request for explicit retry.

Frontend source lives under frontend/todo/. Run npm ci --ignore-scripts, npm test, and npm run build; commit the generated src/chatsite/todo_static/ output with source changes. CI verifies rebuild consistency. Map/chat dependencies are locked and their notices ship with the static assets.

Acceptance covers Python domain/API behavior, the JavaScript controller, real-browser DOM, and real model calls. A successful health response alone is not acceptance: exercise editing, reload persistence, proposals, confirmation, undo and failure paths.