AISB v0.0.1

AISB v0.0.1 packages the current curated AISB benchmark source snapshots from AISB_AUTOSOTA as one zip per benchmark folder for direct download from BenchStore.

What this release contains:

• One zip asset per AISB benchmark package
• The packaged source snapshot, library-authored AGENTS.md, synchronized aisb_catalog.yaml, and current benchmark image / metadata files present in each benchmark directory
• No claim that every benchmark is fully executable on every machine without external datasets, credentials, or upstream evaluation stacks

Usage notes:

• Download the benchmark zip you need from this release or through BenchStore
• Unzip into a local working directory
• Read README.md, AGENTS.md, and aisb_catalog.yaml before attempting reproduction or modification
• Follow each benchmark's dataset, environment, and risk notes carefully; some routes still require external datasets, third-party evaluation repositories, or large hardware budgets

Acknowledgements:

• We thank the original authors of every included paper and codebase for their scientific contributions and open-source releases.
• We also thank the AutoSOTA project for the benchmark collection and organization work: https://github.com/tsinghua-fib-lab/AutoSOTA

v1.5.17

DeepScientist v1.5.17 is primarily about making the path from install to real use more reliable.

If v1.5.16 made Copilot a real first-class mode, v1.5.17 hardens the layers around launch, recovery, Windows compatibility, connector delivery, and operator-facing diagnostics.

The result is a release that feels less fragile at the exact moments that matter most:

• first launch
• onboarding and launch-mode selection
• stalled-turn recovery
• Codex compatibility and bootstrap
• connector delivery and workspace continuation
• local runtime troubleshooting on real machines

Around that core hardening work, this release also expands local-model guidance, adds a real workspace-mode switch in quest settings, improves maintainer docs, and refreshes the public README and startup docs so they match current runtime behavior more closely.

Highlights

1. Launch and onboarding are more reliable

v1.5.17 improves several early-user-path surfaces that were still too easy to confuse or break.

Key changes include:

• fixes for the onboarding launch dialog and demo canvas flow
• improved launch-mode behavior in the workspace and settings surfaces
• a new workspace-mode switch in quest settings
• better startup and continuation handling around the current project shell
• stronger launch and onboarding E2E coverage in the web UI

This matters because the launch path is where DeepScientist has to make its operating model legible immediately.

2. Runtime recovery and direct-turn routing are more disciplined

This release continues the recovery hardening work in the daemon, quest service, and prompt path.

Important changes in this line include:

• better recovery for stalled running turns when new user messages arrive
• follow-up fixes so recovery respects stop and pause correctly
• more stable stale-state handling during recovery watch flows
• refined direct-turn routing behavior
• warmer and clearer style guidance in the interaction path without weakening execution discipline

The practical effect is that DeepScientist is less likely to get stuck in an awkward half-running state and more likely to resume from the current truth cleanly.

3. Windows and local runtime compatibility improved substantially

A major theme in v1.5.17 is making the local runtime behave better on real user machines, especially mixed shell and Windows-adjacent setups.

Notable work in this release includes:

• fixing Windows pop-up / console-window behavior for Codex runner launches
• hiding unwanted runner console windows on Windows
• fixing launcher wrapper home resolution
• sanitizing conda environment state during uv bootstrap
• adding a code-only uninstall flow for cleaner local removal
• tightening launcher and install behavior around managed runtime expectations

This matters because DeepScientist is a local-first product; install and runtime rough edges hurt trust quickly.

4. Connector and Codex compatibility got another important pass

This release also improves the bridge between DeepScientist and the external runtimes or channels it depends on.

Included work in this range:

• fixes across Feishu, Telegram, and WhatsApp connector paths
• Weixin text-send retry handling for stale-context ret=-2 failures
• fixes for MiniMax profile shadowing inside Codex probe homes
• fixes for MiniMax chat-wire MCP tool batching
• better Codex compatibility handling in config, runner, and doctor flows

These changes reduce the gap between “the system is installed” and “the system actually works reliably with the provider or connector I use.”

5. Local model guidance and maintainer docs expanded a lot

v1.5.17 also ships a meaningful documentation step forward.

New or heavily updated areas include:

• a dedicated local model backends guide in both English and Chinese
• updated Quick Start, Doctor, Settings, and Codex provider setup docs
• stronger maintainer-facing development documentation
• README refreshes in both English and Chinese
• clearer startup guidance for non-default homes and local runtime management

This is important because the product surface is broad enough now that missing docs translate directly into user friction.

Detailed change areas

Workspace and web UI

The workspace line received targeted usability and correctness work:

• onboarding launch dialog fixes
• demo runtime / canvas fixes
• workspace mode switch support in quest settings
• file tree and workspace layout improvements
• connector settings form refinements
• stronger client-side API coverage for quest files and session compatibility

Daemon, quest, and launcher runtime

Core runtime layers changed materially in this release:

• daemon app and API updates
• quest service recovery improvements
• launcher and install path fixes
• doctor and diagnostics expansion
• Codex compatibility improvements in runner and config paths
• better bootstrap and runtime-environment handling around local Python / uv / conda

Connectors and provider behavior

The connector/runtime bridge got another stability pass:

• Feishu / Telegram / WhatsApp fixes
• Weixin retry hardening
• MiniMax compatibility fixes
• improved recovery guidance around tool and interaction handling

Docs, README, and release-facing polish

This release also includes broad documentation and repository polish:

• README refresh
• updated startup, doctor, settings, and provider docs
• new local model backend guide
• improved maintainer docs for development and release checks
• npm package @researai/deepscientist@1.5.17 published

Testing and verification

Coverage improved across several layers:

• onboarding and launch-mode E2E tests
• launcher runtime tests
• daemon API tests
• doctor tests
• Codex compatibility tests
• config testing and shared-contract coverage

Why this release matters

v1.5.16 made DeepScientist easier to choose.

v1.5.17 makes it easier to trust after you choose it.

The biggest value in this release is not one single new screen. It is the cumulative reduction in launch friction, runtime ambiguity, recovery fragility, and local-environment surprises.

That is what makes the system feel more like a dependable research operating system instead of a promising but temperamental one.

Version

• npm package: @researai/deepscientist@1.5.17
• npm release: published
• git tag: v1.5.17

v1.5.16

DeepScientist v1.5.16 is primarily about making Copilot real.

If v1.5.15 established DeepScientist as a usable local-first research operating system, v1.5.16 makes it much easier to choose the right operating mode from the start.

The biggest product change is a clean split between:

• Copilot: a quieter, user-directed workspace that waits for your first real instruction
• Autonomous: the standard DeepScientist path that starts moving immediately and keeps the quest advancing on its own

Around that core change, this release also improves Git visibility inside the workspace, reduces local launch friction, fixes important Codex compatibility issues, and substantially expands the documentation around startup, workspace modes, auth, and controller behavior.

Highlights

1. Copilot is now a first-class workspace mode

v1.5.16 introduces a real Copilot workspace mode instead of treating every project start as the same launch style.

What this means in practice:

• you can now create a project that stays idle after creation
• DeepScientist waits for your first real instruction instead of immediately entering the autonomous research loop
• continuation behavior is intentionally more conservative in Copilot mode
• the distinction is now reflected in startup UX, prompts, docs, and runtime behavior

This is the most important change in the release.

2. New launch-mode split: Copilot vs Autonomous

Project start now explicitly separates launch intent.

Instead of a single undifferentiated creation path, DeepScientist now exposes a clearer choice between:

• Copilot Mode
• Autonomous Mode

This changes more than wording. It affects:

• whether the quest starts running immediately
• whether the system parks after the current requested unit
• how continuation should behave
• how much autonomous routing is appropriate

New docs explain this split directly:

• docs/en/20_WORKSPACE_MODES_GUIDE.md
• docs/zh/20_WORKSPACE_MODES_GUIDE.md

3. Dedicated Copilot project creation flow

v1.5.16 adds a dedicated Copilot creation flow with a much lower-friction path into an interactive workspace.

New Copilot creation capabilities include:

• starter card selection
• optional connector binding during creation
• live preview of the selected project display style
• card background and accent customization
• clearer “idle by default” framing

This makes Copilot much more usable for repo inspection, paper reading, debugging, experiment planning, and request-scoped collaboration.

4. Git visibility inside the workspace is much stronger

This release improves the Git-facing workspace experience substantially.

New and improved capabilities include:

• Git commit viewer support inside the plugin/workspace system
• better commit-level diff inspection
• snapshot vs diff switching while inspecting commit changes
• stronger Git operation visibility in lab/canvas-related surfaces
• better Git-oriented artifact and projection handling

This matters because Git is increasingly part of durable research state, not just an implementation detail.

5. Local browser auth is now optional and disabled by default

A notable behavior change in v1.5.16:

• local browser auth is now optional
• default startup no longer requires the local browser password gate
• if you want the password gate, start with ds --auth true

This lowers friction for normal local-first usage while preserving an explicit browser gate when users want it.

Related docs:

• docs/en/19_LOCAL_BROWSER_AUTH.md
• docs/zh/19_LOCAL_BROWSER_AUTH.md

6. Codex compatibility and runtime behavior improved

This release includes several important Codex-side fixes:

• MiniMax Codex compatibility was repaired
• Codex profile handling and model: inherit guidance were improved
• the Codex runner runtime-home semantics were corrected so DeepScientist no longer drops existing user Codex customizations on the same machine

A particularly important fix in the v1.5.16 line:

• DeepScientist now runs Codex from an isolated .ds/codex-home
• that runtime home first inherits the configured real Codex home semantics
• this preserves existing auth, config, skills, agents, and prompts
• quest-local .codex prompts/skills are then overlaid on top

This fixes a real mismatch where plain terminal codex and DeepScientist-launched codex could previously see different environments.

Detailed change areas

Copilot and workspace behavior

Major work landed across the workspace surface, project creation flow, shell layout, and runtime semantics:

• explicit Copilot vs Autonomous startup split
• new CreateCopilotProjectPage
• new launch-mode visuals and launch-mode dialogs
• improved Copilot composer / dock / pane layout behavior
• better session compatibility handling in the UI
• improved event/session restoration behavior
• better Copilot-facing onboarding and create-flow copy
• E2E coverage for Copilot workspace flows

Projects / landing / startup UX

The launch and projects surfaces received a broad refresh:

• refreshed landing / hero flow
• improved Start Research and launch-mode selection experience
• stronger project preview cards
• better onboarding flow wording and step coverage
• improved project opening and display preview behavior

Git / canvas / explorer improvements

Work in this release continues the shift toward Git as a visible part of the research operating system:

• Git commit viewer plugin
• improved Git diff viewer behavior
• stronger Git projection support in artifacts and quest state
• improved branch/canvas-related API and tests
• additional Git operation event coverage

Runtime / daemon / API

Several backend layers changed materially:

• daemon API and app changes
• artifact service expansion
• quest service and stage view updates
• bash_exec service updates
• prompt builder changes
• Copilot-specific prompt path via system_copilot.md
• session and ACP compatibility improvements
• more test coverage across daemon API, MCP, prompts, git branch canvas, and Codex runner behavior

Docs and release polish

Documentation changed heavily in this release and is much more aligned with current runtime behavior.

New or significantly updated topics include:

• workspace modes
• local browser auth
• external controller guide
• quick start
• settings reference
• Codex provider setup
• runtime and canvas behavior
• prompt / skills / MCP behavior
• README refresh in both English and Chinese

Packaging and release-side polish also improved:

• better release build handling
• updated ignore/package boundaries
• published npm package for 1.5.16

Upgrade notes

Browser auth behavior changed

If you were used to the local browser password flow, note that:

• auth is now disabled by default
• to enable it explicitly, use:

ds --auth true

Copilot is now a separate startup path

If you want a quiet project shell that waits for your first instruction, choose Copilot.

If you want DeepScientist to immediately start converting the startup contract into real work, choose Autonomous.

Codex runtime-home semantics changed

For Codex users, the runtime-home model is now cleaner:

• quest-local .codex/ remains the quest overlay location
• actual runtime Codex home now lives under .ds/codex-home

This should improve compatibility, especially for users with an existing global Codex setup.

Why this release matters

v1.5.15 already had the broad DeepScientist shape.

v1.5.16 makes that shape much more practical.

The most important product difference is that DeepScientist now has a clearer answer to two very different ways people want to work:

• “Stay quiet, wait for me, help me one unit at a time”
• “Take the quest and keep moving”

That split is what makes the new Copilot path matter.

Around that core change, this release also tightens the runtime, improves Git inspection, reduces local launch friction, and brings the docs much closer to the actual system.

Version

• npm package: @researai/deepscientist@1.5.16
• npm release: already published
• git tag: v1.5.16

DeepScientist v1.5.15

Overview

DeepScientist v1.5.15 is a broad maintenance and workflow-hardening release.

Compared with v1.5.14, this release does not only adjust one metrics view. It also updates the prompt and skill contracts, hardens bootstrap and stage routing, improves connector/runtime behavior, expands compatibility, and refreshes several user-facing surfaces.

Major Functional Changes

1. Baseline comparison and metrics surfaces

This release reworks how baseline information is exposed in the quest workspace.

Changes:

• Metrics Overview is now focused on the current active baseline plus main-experiment traces.
• Baseline Compare is separated out as its own comparison surface for confirmed baselines and variants.
• The internal metric builders, quest service, daemon API, and web UI were updated to support this split.
• A dedicated quest-level compare route is now exposed for baseline comparison data.

Why this matters:

• it is clearer which baseline the current experiment line is being judged against
• it is easier to inspect other confirmed baselines and variants without mixing them into the active timeline
• the workspace is less likely to blur "current comparator" and "all known baselines"

2. Baseline confirmation now preserves richer metric contracts

This release strengthens the baseline contract rules so a baseline is less likely to be reduced to only one headline scalar when richer comparable metrics already exist.

Changes:

• artifact.confirm_baseline(...) is now backed by stronger metric-surface handling.
• Prompt and skill rules now explicitly say that primary_metric is only the headline metric.
• Multi-task, multi-split, multi-variant, or multi-metric baseline surfaces should be preserved in metrics_summary, metric_contract, and baseline_variants.
• Existing structured baseline-side contracts should be reused instead of being replaced by a thinner hand-written summary when possible.

Why this matters:

• quests with richer baseline packages keep more of the comparison information needed later
• users can compare variants and task-level performance more accurately
• future runs are less likely to silently optimize toward an over-compressed baseline target

3. Connector-facing interaction messages keep the full answer

This release separates the real user-facing interaction body from the preview/summary layer.

Changes:

• actual connector-facing interaction text is preserved in full
• preview text is generated separately for compact displays and summaries
• artifact interaction journaling records the full message and the preview layer distinctly
• the system prompt now explicitly warns agents not to manually turn the real user message into a shortened preview

Why this matters:

• connector users are less likely to receive milestone or baseline messages that were already truncated with …
• journaling and delivery behavior are easier to reason about
• user-facing updates can stay readable without throwing away decision-critical detail

4. Prompt, SOP, and execution-discipline updates

This release also continues the recent prompt and skill hardening work.

Changes in this version line include:

• stronger SOP and workflow guidance in the system prompt
• clearer bash_exec-first terminal execution rules
• stronger prompt redlines against using the wrong shell path
• clearer baseline-discipline language in the baseline skill and prompt/MCP docs

Why this matters:

• stage behavior is easier to keep aligned with the intended workflow
• future quests are less likely to drift into shell usage or baseline handling that conflicts with system policy
• maintainers have a clearer prompt/skill contract to inspect when workflow quality regresses

5. Bootstrap, routing, and runtime behavior

This release contains several fixes around early quest routing and runtime stability.

Changes:

• bootstrap quest stage routing is more stable
• install path and quest lock import behavior were hardened
• connector/workflow/runtime behavior was refreshed in the broader runtime stack
• baseline lineage and connector delivery behavior received additional hardening

Why this matters:

• new quests are less likely to drift into the wrong stage too early
• route changes around baseline and experiment setup are less fragile
• runtime state is more predictable during startup and resume flows

6. Connector and platform improvements

This version line also includes broader compatibility and connector updates.

Included work in this range:

• Windows compatibility improvements
• connector expansion work for additional connector surfaces
• Weixin delivery retry improvements for the ret=-2 text-send failure class
• connector delivery and workflow behavior updates across the stack

Why this matters:

• more environments can run the product cleanly
• connector delivery is more resilient
• the release is not only about the web UI; it also improves operator-facing delivery paths

7. User-facing UI and content fixes

This release also bundles several user-visible fixes and cleanup items.

Included work in this range:

• improved quest loading behavior
• tutorial locale alignment fixes so tutorial content stays in the chosen UI language more consistently
• landing hero interaction stabilization
• removal of temporary internal maintenance files and image artifacts from the public tree

Why this matters:

• startup and workspace loading are more coherent
• language-switch behavior is less surprising
• the public repository and packaged output are cleaner

User-Visible Impact

If you are using the web workspace, the most visible changes are:

• clearer separation between active baseline tracking and full baseline comparison
• better retention of detail in connector-facing milestone messages
• improved loading and routing behavior for quests
• cleaner language behavior in tutorial-related UI flows

If you are maintaining prompts, skills, or quest workflows, the most important changes are:

• stronger baseline confirmation rules in the system prompt
• stronger bash_exec execution discipline
• more explicit prompt/skill guidance for preserving rich comparison contracts
• more stable bootstrap routing behavior

Upgrade Notes

• Old quests that already stored only one aggregate baseline metric will still show only that aggregate metric until a richer baseline contract is recorded for that quest.
• The npm package for this release is published as @researai/deepscientist@1.5.15.
• This release includes both behavior changes and contract hardening, so prompt/skill consumers should read the updated system prompt and baseline skill expectations rather than assuming only UI changes landed.

Commit Themes Since v1.5.14

Representative changes in this release line include:

• feat: split baseline compare and interaction previews
• fix: stabilize stage routing for bootstrap quests
• fix: harden install path and quest lock import
• feat: front-load bash_exec redlines in system prompt
• feat: enforce bash_exec-only terminal execution
• feat: strengthen prompt SOP and metrics overview surfaces
• Update connector, workflow, and runtime behavior
• Default Codex to yolo and unblock built-in MCP tools
• fix: stabilize landing hero interactions
• fix: keep tutorial language aligned with current ui locale
• feat: improve quest loading and release v1.5.15
• 兼容 windows，修复中英文翻译，新增 connector
• Harden baseline lineage and connector delivery
• fix: retry weixin text sends after ret=-2

DeepScientist v1.5.14

Overview

DeepScientist v1.5.14 focuses on runtime reliability, workflow updates, and UI/TUI polish across the local-first research loop.

What Changed

• Versioning and packaging updated to 1.5.14 across npm and Python manifests (package.json, pyproject.toml, src/ui/package.json, src/tui/package.json, and lockfiles).
• Runtime/backend updates across artifact, daemon, MCP, and quest services:
  • src/deepscientist/artifact/service.py
  • src/deepscientist/daemon/app.py
  • src/deepscientist/mcp/server.py
  • src/deepscientist/quest/service.py
  • src/deepscientist/quest/stage_views.py
  • src/deepscientist/prompts/builder.py
• Prompt and skill workflow refinements:
  • src/prompts/system.md
  • src/skills/baseline/SKILL.md
  • baseline references under src/skills/baseline/references/
• TUI improvements (including connector-facing logic and API/type updates):
  • new src/tui/src/lib/connectors.ts
  • updates in src/tui/src/app, src/tui/src/components, src/tui/src/lib/api.ts, src/tui/src/types.ts
• Web workspace/lab updates:
  • src/ui/src/components/workspace/*
  • src/ui/src/lib/plugins/lab/components/LabQuestGraphCanvas.tsx
  • related UI API/store updates
• Documentation updates in English and Chinese TUI/docs index pages.
• Test updates across daemon API, MCP servers, memory/artifact, prompt builder, skill contracts, and stage skills.

Release Artifacts

• npm package published: @researai/deepscientist@1.5.14

DeepScientist v1.5.12

DeepScientist v1.5.12

This release is published as tag v1.5.12 and corresponds to npm package @researai/deepscientist@1.5.12.

Compared with v1.5.11, this update focuses on workflow polish, paper-oriented review and rebuttal support, canvas clarity, LaTeX stability, memory safety, and public npm release readiness.

Highlights

Start Research and custom launch workflow

• Reworked Start Research custom-mode inputs for review, rebuttal, revision, and continue-existing-state scenarios.
• Replaced the launch-mode dropdown with card-style launch-path selection and matching entry-type cards underneath.
• Clarified the standard vs custom entry structure so the UI reads more like a real workflow contract instead of a loose form.
• Removed the explicit tutorial-example fill button from Start Research.
• Kept support for both web links and local file/folder paths in research, review, and manuscript inputs.

Review, rebuttal, and paper experiment planning

• Strengthened the built-in review and rebuttal skills so they can plan and route reviewer-driven follow-up work more explicitly.
• Added a paper experiment matrix contract for paper-facing supplementary experiments.
• The writing / analysis / review / rebuttal flow now treats the paper experiment matrix as the durable source of truth for:
  • required main-text follow-up experiments
  • appendix-only experiments
  • optional case studies
  • highlight-validation experiments
  • efficiency / cost experiments when relevant
• Added a reusable paper-experiment-matrix reference template.

Canvas, onboarding, and current-path clarity

• Added the unified connector + guided demo entry coach dialog.
• Improved workspace canvas clarity so the current branch / current path is easier to see.
• Added current-path highlighting and stronger current-node emphasis in the lab canvas.
• Improved the first-run guidance flow around connector binding and demo entry.

LaTeX and file-opening stability

• Fixed the notebook-style socket client compatibility issue around socket.once.
• Fixed the LaTeX editor on quest runtime surfaces so it no longer depends on a Socket.IO channel that the quest runtime does not expose.
• Added a local-file fallback mode for LaTeX editing on the quest runtime.
• Improved the LaTeX PDF preview fitting behavior so the rendered page is more likely to fit the pane cleanly by default.

Runtime safety and large-quest stability

• Reduced memory blowups caused by oversized JSONL event lines, especially around large tool-result payloads.
• Added safer streaming / preview behavior for large bash_exec logs and quest event reads.
• Added quest-slimming and cold-file dedupe helper scripts for oversized quest worktrees.
• Hardened several quest-runtime and channel paths that previously loaded too much data eagerly.

Packaging, provider setup, and public npm release path

• Added Codex provider setup docs in both English and Chinese.
• Bumped the project, Python package, UI, and TUI versions to 1.5.12.
• Hardened the release build path for the public npm package.
• Updated the UI build script so release prebuilds do not depend on missing shell shims for vite in this environment.
• Updated the release script so npm pack / npm publish can rely on prebuilt bundle freshness instead of recursively rebuilding in a broken prepack path.

Included areas of change

• Launcher and packaging: bin/ds.js, package.json, package-lock.json, pyproject.toml, scripts/build-release.mjs
• Runtime and daemon: src/deepscientist/daemon/, src/deepscientist/bash_exec/, src/deepscientist/quest/, src/deepscientist/mcp/
• Prompts and skills: src/prompts/, src/skills/
• Web UI: src/ui/src/components/, src/ui/src/lib/
• Docs: docs/en/, docs/zh/
• Tests: multiple prompt, skill, runtime, connector, and UI-facing tests

Release notes

• Git tag: v1.5.12
• npm package: @researai/deepscientist@1.5.12
• Compare: v1.5.11...v1.5.12

DeepScientist v1.5 Stable Release

DeepScientist v1.5 Stable Release

This release is published as tag v1.5.11 and marks the stable public release of DeepScientist v1.5.

DeepScientist is not just a long-running autonomous scientific discovery system. It is also a persistent research map that lives on your own machine.

Highlights

• A local-first research operating system for verifiable computational research.
• A persistent research map built from Git branches, artifacts, memory, connector traffic, and quest events.
• A workshop-style workflow: DeepScientist can drive end to end, while humans can pause, inspect, edit code, and run terminal commands at any time.
• A release-oriented npm package with a Python-first runtime and prebuilt Web and TUI bundles.
• One bound external connector per quest, with current stable support centered on QQ, Weixin, and Rokid Glasses.

What is included in v1.5

Core runtime and research workflow

• Added the current DeepScientist launcher, local runtime bootstrap, uv-managed Python environment setup, doctor flow, and migration flow.
• Added the quest runtime with the core contract of one quest as one Git repository.
• Added durable quest state, stage views, node traces, Git-backed workspace flows, and branch-aware research structure.
• Added prompt-led and skill-led workflow control, including the system prompt, prompt builder, and first-party research skills.
• Added the built-in MCP workflow centered on memory, artifact, and bash_exec.

Persistent research map

• Added the current Studio / Canvas style workspace for tracking branches, experiments, files, notes, and execution state.
• Added live project and quest surfaces that reconstruct progress from durable runtime state instead of collapsing work into a single chat log.
• Added memory, artifact guidance, metrics, and reusable baseline flows so later rounds can build on previous results instead of restarting from zero.

Workshop-style interfaces

• Added the shared Web workspace and TUI experience on top of the same daemon API and event model.
• Added file exploration, project views, notebook and markdown flows, PDF and LaTeX-related surfaces, and workspace navigation improvements.
• Added onboarding and guided workflow flows so users can understand how to start and manage a research loop inside DeepScientist.

Connectors

• Added the channel framework covering QQ, Weixin, and Rokid Glasses.
• Added QQ connector support with prompt and bridge handling aligned to quest delivery.
• Added Weixin QR-based binding and delivery flow, with corresponding setup and usage guides.
• Added Lingzhu / Rokid Glasses support, including generated binding fields, simplified setup flow, and connector guide.
• Added connector-aware prompt and delivery behavior so quests can use the selected external communication path cleanly.

Documentation

• Added a complete English and Chinese documentation structure for the current product.
• Added Quick Start, Settings Reference, Start Research Guide, TUI Guide, Runtime and Canvas guide, Memory and MCP guide, Weixin guide, QQ guide, Lingzhu / Rokid guide, Guided Workflow Tour, Core Architecture Guide, Prompt / Skills / MCP Guide, acknowledgements, and License / Risk documentation.
• Updated the main README to reflect the current DeepScientist positioning, workshop-style collaboration model, connector support, and install path.

Packaging and release

• Added release-oriented packaging for @researai/deepscientist.
• Added prebuilt Web and TUI bundles for public npm installation.
• Added install and deployment scripts aligned with the current launcher and runtime structure.
• Released npm package version 1.5.11 in sync with this GitHub release.

Contribution workflow

• Added CONTRIBUTING.md.
• Added GitHub issue templates for bug reports, features, docs, and RFCs.
• Added a pull request template and repository hygiene updates for public contribution.

Testing and validation

• Added extensive tests across launcher, daemon API, quest runtime, MCP, connectors, memory, artifact flows, Git graph behavior, and workspace surfaces.
• Added end-to-end scripts and fixtures for validating workspace and lab flows.

Notes

• This release should be treated as the stable public baseline for DeepScientist v1.5.
• The Git tag is v1.5.11, while the release message is intentionally presented as the stable release of the v1.5 line.
• Future releases can be described as incremental updates on top of this public baseline.