v0.2.5 — OpenClaw Auto-Detection, Init Rewrite, Server Overhaul

This release transforms the first-run experience: taskmeld init now auto-detects a local OpenClaw installation, offers one-click install if missing, and guides users through Gateway configuration — no more manual setup. The server startup is rewritten to run foreground-first with automatic local Gateway lifecycle management.

Added

• OpenClaw local detection and management — New src/openclaw/ module:
  • openclaw-detector.ts: Detect local OpenClaw installation, version, and config path
  • openclaw-installer.ts: One-click install via npm install -g openclaw@latest
  • openclaw-gateway.ts: Gateway start/stop/status management and health polling
  • openclaw-types.ts: Type definitions for detection results and config schema
• Init orchestrator (src/init/): Rewritten interactive setup flow with branching logic:
  • Auto-detect local OpenClaw → auto-connect if running with token
  • Three paths on missing install: auto-install, remote Gateway, manual
  • Optional model configuration delegation to openclaw configure
• Non-interactive init modes: --auto-install, --remote <url>, --skip-model-config flags for CI/CD
• Server runtime client: Foreground-first server startup with startup lock, health polling, and ownership probing
• 10 new test files (~1400 lines) covering Wevra loop/executor/memory/registry/permissions, pipeline readiness, OpenClaw detector, CLI enhanced init, and LLM client stream

Changed

• Server start rewrite: taskmeld server start now runs in foreground; auto-starts local OpenClaw Gateway when installed locally
• Init command rewrite: From simple Gateway connection prompt to full orchestration with detection, installation, and branching user choices
• UI prompts: Added numberedSelectPrompt, maskedFieldPrompt for init flow
• Localization: New translation keys for init flow (en/zh.json +61 each)
• README overhaul: Updated architecture diagram, prerequisites, install flow, quick start, directory structure, and roadmap to reflect OpenClaw management capabilities
• Version bumped to 0.2.5

Fixed

• Gateway connect errors: Retryable connection failures treated as transport errors, properly propagated up
• Init remote config: Remote Gateway URL/Token saved without requiring live connection validation
• Security: API and Vite dev server restricted to 127.0.0.1 only, no longer binding to 0.0.0.0
• Server process group: Cleaner daemon lifecycle management

Full Changelog

v0.2.4...v0.2.5

v0.2.4 — Openclaw Gateway Agent/Skill/Session Tools Integration

What's Changed

Added

• Openclaw Gateway agent/skill/session tools — Wevra Agent now manages Openclaw agents through the gateway instead of local abstractions
  • gateway_agent_list, gateway_agent_create, gateway_agent_update, gateway_agent_delete, gateway_agent_send tools
  • gateway_session_list, gateway_session_get, gateway_session_history tools
  • gateway_skill_install, gateway_skill_update, gateway_skill_search, gateway_skill_status tools
  • skills.install RPC supporting clawhub/upload/installer modes
  • pipeline_node tool validates agentId references against registered Openclaw agents
  • Gateway RPC discovery documentation (docs/gateway-rpc-discovery.md)

Changed

• Agent tools architecture — removed built-in agent.ts (353 lines), replaced with gateway.ts (588 lines) routing all agent operations through Openclaw Gateway
• Agent loop — removed hard maxIterations=25 cap; loop terminates naturally when LLM produces no tool calls
• Model config — inherit contextWindow/maxTokens from builtin providers when missing from persisted config; getModelsConfigPublic() returns templates as single source of truth
• Frontend model config — ModelConfigModal uses backend templates instead of duplicated provider constants
• Deprecated models removed — dropped Xiaomi MiMo V2 Flash, MiMo V2 Pro, MiMo V2 Omni

Fixed

• Typecheck script — generate version.ts before running tsc to prevent missing module errors

---

Full Changelog: v0.2.3...v0.2.4

v0.2.3 - Blueprint Visualization & Gateway Reconnection Fix

What's New

Blueprint Preview and Visualization

• Interactive flow diagram for pipeline blueprints
• blueprint tool for Wevra Agent to generate and visualize pipelines
• BlueprintPreviewPanel with sidebar drawer toggle and fullscreen mode
• BlueprintFlow component with interactive node graph and dependency visualization
• BlueprintNodeCard component for individual node rendering
• Automatic layout algorithm based on dependency graph and routing paths
• Wheel zoom support for flow diagram navigation

Bug Fixes

Gateway Reconnection

• App layer now detects successful reconnections and refreshes runtime state
• gateway-client: Store lastHello payload and expose getHello() getter
• create-app-context: Detect status→ready transition after first connect, replay onGatewayReady(hello) to refresh runtime state and broadcast to frontends
• index.ts: Log gateway-reconnected / gateway-status failures for observability

v0.2.2 — ask_user Structured Question Tool & Pipeline Design Protocol

Highlights

🗣️ ask_user Structured Question Tool

Wevra Agent can now ask users structured questions with predefined options instead of free-form text. Supports multi-question batches with tab navigation, per-tab "Other" custom input, single-select auto-advance, and full answer context (question + option label + description) returned to the LLM.

📋 Pipeline Design Protocol

The pipeline-management skill expanded from 6 lines to a comprehensive 5-phase protocol guiding Wevra through reliable pipeline creation: Goal Discovery → Architecture Proposal → Agent Assignment → Artifact Design → Incremental Build. Includes 26 red lines and artifact content strategy.

What's Changed

Added

• ask_user tool with multi-question batch, per-tab Other input, auto-advance, skip all
• InlineQuestion inline component (replaces QuestionDialog modal)
• ChatInputArea component extracted from WevraChatPanel
• wevra.ask-user WebSocket method and question_request/question_response stream events
• Pipeline Design Protocol with red lines, artifact strategy, anti-patterns
• ResultEnvelope documentation in pipeline_node tool description
• Question state persistence via localStorage (survives page refresh)

Fixed

• Question timeout crash (conversation no longer breaks when user doesn't answer in time)
• Stream events lost after answering a question (LLM response now syncs correctly)
• Page refresh stuck in busy state (pending question restored from localStorage)
• Other option state leaking across tabs
• Other button not responding on first tab
• Confirm button initial state (starts disabled)
• Textarea not hiding when switching from Other to a regular option

Changed

• Question timeout increased to 10 minutes
• Input area extracted into standalone ChatInputArea component
• pipeline-management skill: 6 lines → full 5-phase protocol

Full Changelog: v0.2.1...v0.2.2

v0.2.1 — Per-conversation model persistence & model management redesign

What's Changed

🐛 Fixed

• Model switching not working — selecting a different model now takes effect immediately and persists across sessions
• "No LLM model configured" error — server no longer requires restart after adding providers via UI
• Thinking level race condition — eliminated concurrent state overwrite when opening the Wevra panel
• Context window not updating — switching models now correctly updates the context usage ring
• New conversation shows "model" placeholder — displays the configured default model name

✨ Changed

• Per-conversation model storage — each conversation remembers its selected model, restored on switch
• Model management modal redesign
  • "Add Provider" tab shown first with DeepSeek pre-selected
  • "Models" tab shows all configured models grouped by provider
  • Radio button to set default model
  • Toggle slider to enable/disable models (disabled models are hidden from the chat dropdown and cannot be called)
  • Disabling the default model auto-switches to the next available
• Auto-open model config — modal pops up automatically when no models are configured on first launch
• New backend APIs — wevra.models.enable / wevra.models.disable WebSocket methods
• Default thinking level — new conversations initialize with high instead of medium

v0.2.0 — Wevra Agent: Built-in AI assistant for pipeline management

🚀 Highlights

Wevra is a built-in AI agent that lets you manage pipelines through natural language. Ask it to create, inspect, diagnose, or control pipelines — it calls the right tools and reports back.

✨ Added

Wevra Agent

• ReAct (Reasoning + Acting) loop for intelligent task execution
• 28 built-in tools covering Pipeline, Agent, System, Memory, Skill, Artifact, Session, and Web operations
• Multi-provider LLM support: DeepSeek, OpenAI, Xiaomi MiMo, and custom providers
• Real-time streaming of thinking process and execution results via WebSocket
• Three permission modes: Plan (read-only), Normal (confirm writes), Auto (full access)
• Cross-session memory system for knowledge accumulation
• Skill system with 3 built-in skills: core-behavior, pipeline-management, failure-diagnosis
• Conversation management with JSONL persistence and crash recovery

Agent Tools (6 tools)

• agent_list, agent_get, agent_create, agent_update, agent_delete, agent_send
• Full agent lifecycle management with confirmation dialogs

Pipeline Tools (12 tools)

• Pipeline CRUD, run/stop, diagnosis, plugin management, node management, pre-run validation

Web Console

• WevraChatPanel for natural language interaction
• Message bubbles, conversation sidebar, mode switching, confirmation dialogs
• Model selection, context usage bar, debug panel

WebSocket API

• 19 methods: wevra.chat, wevra.conversations.*, wevra.models.*, wevra.config.*, wevra.tool-preferences.*, wevra.confirm

🔧 Technical

• Zero-dependency Agent framework using native fetch for OpenAI-compatible APIs
• ~350 lines of code for the entire reasoning layer
• Tool-based architecture: LLM calls tools, tools call services
• Frozen prompt mechanism for cache efficiency
• Per-conversation mode tracking and preference management

v0.1.50 — Agent CRUD Management

Added

• CLI: agent create, agent update, agent delete commands for agent
  lifecycle management
• Web: Create, edit, delete agent modals with action buttons in agent
  list
• Backend: agent.create, agent.update, agent.delete,
  agent.defaultWorkspace WebSocket methods
• Auto-resolve default workspace path on agent creation (env →
  config.json → relative fallback)
• Auto-detect and persist workspace root from gateway config on connect

Fixed

• Hardcoded English error messages in agent modals replaced with i18n
  keys
• Escape key not closing agent modals
• Edit agent modal now pre-populates current name and workspace values

v0.1.41 — i18n Internationalization & Settings

Added

• i18n infrastructure with i18next and locale files (en, zh)
• Settings page with language selector, English as default
• All frontend widgets and pages migrated to translation keys
• CLI commands migrated to use translation keys
• Backend error/info messages replaced with English equivalents

Changed

• Navigation labels decoupled from routing keys for i18n flexibility
• 33 audit issues resolved across i18n implementation (DRY, naming,
  structure)

Fixed

• Server static file serving path for web dist directory

v0.1.3 — WebSocket Transport Migration

Breaking Changes

• Migrate all API communication from HTTP REST to WebSocket transport
• Remove HTTP route-based architecture (router, middleware, route
  handlers)

Added

• WebSocket method system (src/transport/ws-methods/) with dedicated
  handlers for pipelines, agents, sessions, artifacts, logs, timeline
• ws-handler.ts for unified WebSocket request routing
• Frontend WebSocket client (ws-client.ts) replacing HTTP-based API
  layer
• CLI WebSocket runtime client (ws-runtime-client.ts)
• Isolated dev data directory to prevent cross-environment conflicts

Changed

• All frontend entity APIs (pipeline, agent, session, artifact, run-log)
  migrated to WebSocket calls
• CLI server runtime client refactored to use WebSocket transport
• Pipeline management tests updated for new transport layer

Removed

• HTTP router, middleware, and all REST route handlers (17 route files)
• Server-side HTTP utility modules
• Legacy server contract and route tests

v0.1.2 — Initial Release

v0.1.2 — Initial Release

The first public release of TaskMeld, an agent pipeline orchestration platform for OpenClaw.

What's included

• CLI tool taskmeld — pipeline, agent, server, scheduler, artifact, system
• taskmeld init — interactive guided setup for OpenClaw Gateway connection
• HTTP REST API — full control plane with Trie-based routing
• WebSocket broadcast — real-time pipeline and gateway state streaming
• React console — DAG visualization, agent sessions, artifact browser, log viewer
• DAG pipeline engine — parallel groups, routing branches, per-node retry, state machine
• OpenClaw Gateway integration — WebSocket client for auth, event relay, agent session delegation
• User-level config and data — ~/.taskmeld/, zero external database dependency

Requirements

• Node ≥ 18
• OpenClaw ≥ 5.20
• Windows (macOS and Linux are not yet tested)

Install

npm install -g taskmeld
taskmeld init
taskmeld server start