v0.5.0 — Autopilot Mode

Autopilot Mode for Blueprints

ClawUI v0.5.0 introduces Autopilot mode — an autonomous agent loop that executes blueprint nodes without manual intervention.

Highlights

• Autopilot execution engine — autonomous loop that picks the next ready node, executes it, and advances the blueprint toward completion
• 19-tool decision palette — the autopilot agent chooses from tools like execute-node, enrich-node, split-node, smart-deps, skip-node, reevaluate-all, and more
• Infinite-loop safeguards — iteration limits, stuck-detection heuristics, and automatic pausing when progress stalls
• Memory & reflection system — autopilot learns from each iteration, building context that improves decisions across the execution loop
• Autopilot UI — toggle switch to enable/disable, real-time decision log, pause banner with resume capability
• plan-operations.ts extraction — shared AI operation logic (enrich, reevaluate, split, smart-deps) used by both route handlers and autopilot
• Execution history tracking — full audit trail of autopilot decisions and tool invocations

Install

npm install -g @clawui/cli && claw-ui
# or
npx @clawui/cli

Full Changelog

ClawUI v0.4.0

ClawUI v0.4.0 Release Notes

ClawUI v0.4.0 is a major release that transforms ClawUI from a Claude Code-specific tool into a multi-agent session manager. This release adds support for four agent runtimes, full Windows compatibility, a role-based blueprint system, and an AI-powered insights mechanism.

---

Highlights

Multi-Agent Support

ClawUI now supports 4 agent runtimes as backends:

• Claude Code — the original, fully supported runtime
• Codex CLI — OpenAI's Codex CLI with session parsing and sandbox-bypass handling
• OpenClaw — including Docker multi-directory support with OPENCLAW_PROFILE
• Pi Mono — lightweight alternative runtime

Switch between runtimes with the AGENT_TYPE environment variable. Each runtime has full JSONL parsing, session sync, health analysis, and blueprint execution support. The UI is now agent-neutral — all user-facing text refers to "agent" rather than a specific product name.

Windows Support

Full cross-platform compatibility across all features:

• All agent runtimes resolve binaries on Windows (.cmd shims in AppData/Roaming/npm/, where.exe PATH lookup)
• Cross-platform path encoding handles backslashes and drive letter colons
• Windows-compatible temp directories and shell execution
• CI matrix now includes Ubuntu + Windows on Node 20 + 22

Multi-Role Blueprint System

Assign specialized roles to blueprint nodes for more targeted AI execution:

• SDE (blue) — Software Development Engineer tasks
• QA (green) — Quality Assurance and testing tasks
• PM (purple) — Product Management and planning tasks

Role-aware prompt assembly tailors agent instructions per role. Visual RoleBadge and RoleSelector components with semantic color coding. When no roles are configured, the system defaults to SDE-only mode (fully backward compatible).

Blueprint Insights

An automatic insight system that surfaces actionable intelligence during blueprint execution:

• Insights are generated during node execution and evaluation
• Three severity levels: info, warning, critical — each with distinct color coding
• NavBar shows an unread count badge on the Blueprints link
• The Plan Coordinator reads unread insights and suggests blueprint graph changes (add/update nodes, dismiss low-value insights)

---

New Features

• Multi-agent runtime architecture with pluggable AgentRuntime interface — all blueprint operations (generation, execution, evaluation, enrichment) route through the active runtime
• Codex CLI integration with session parsing, --dangerously-bypass-approvals-and-sandbox support, and ~/.codex/config.toml trust requirement handling
• OpenClaw Docker support — scan local + Docker instance session directories via OPENCLAW_PROFILE environment variable
• Role registry system with SDE, QA, and PM built-in roles, resolveNodeRoles() and buildArtifactPrompt() helpers
• Role API endpoints — GET /api/roles and GET /api/roles/:id
• Blueprint Insights table with full API: callbacks, listing, mark-read, mark-all-read, dismiss, and global unread count
• Plan Coordinator for insight-driven blueprint graph management
• Cross-tab state sync via BroadcastChannel — other open tabs auto-refresh when operations start or sessions run
• Per-session run lock — prevents concurrent resume processes on the same session (HTTP 409 on conflict)
• Toast notification system — lightweight custom implementation with auto-dismiss, progress bar, and success/error variants
• Dark/light mode toggle with CSS variable-based theming and semantic color tokens
• Context-aware tooltips on all AI-triggered buttons (enrich, reevaluate, smart deps, split)
• 2-line description preview on blueprint node cards with show more/less toggle
• Session live-polling — 2s poll interval during active runs for faster response streaming
• Dev/stable environment separation — isolated ports and databases for parallel development and production use
• Dev redeploy endpoint — one-click stable redeployment from the dev UI
• Global status endpoint — GET /api/global-status for aggregate queue info across all blueprints
• Comprehensive test suite — 907 backend tests across 22 files + 416 frontend tests across 28 files, including Windows platform coverage

---

Bug Fixes

• Auto-finalize blueprint — blueprints now correctly transition to done status when all nodes reach done or skipped via API callbacks
• MarkdownContent rendering — fixed rendering issues in the markdown content component
• Windows server startup — index.ts adjusted for Windows platform compatibility
• Windows path handling in routes — route handlers updated for cross-platform path operations
• Cross-platform path encoding — encodeProjectCwd now uses /[/\\]/g regex and handles drive letter colons
• Missing logger instance in agent-claude.ts
• Post-merge type errors, lint errors, and test failures resolved
• 21 performance and security issues identified and fixed from codebase audit
• 14 additional bugs fixed across backend and frontend

---

Breaking Changes

None. Existing setups continue working without changes. All new features are additive and backward compatible.

---

Upgrade Guide

Install

npm install -g @clawui/[email protected]

Or run directly:

npx @clawui/[email protected]

New Environment Variables (all optional)

Variable	Description	Default
AGENT_TYPE	Agent runtime: claude, openclaw, pi, codex	claude
OPENCLAW_PATH	Path to OpenClaw CLI binary	Auto-detected
PI_PATH	Path to Pi Mono CLI binary	Auto-detected
CODEX_PATH	Path to Codex CLI binary	Auto-detected
OPENCLAW_PROFILE	OpenClaw Docker profile name	Unset
CLAWUI_DEV	Enable dev mode (reuse auth token, dev UI features)	Unset

Backward Compatibility

• Existing setups continue working unchanged — AGENT_TYPE defaults to claude
• Role system defaults to SDE-only mode when no roles are configured on a blueprint
• Insights are generated automatically during execution — no configuration needed
• Windows users benefit immediately from cross-platform path resolution

For Windows Users

ClawUI now fully supports Windows. Agent CLI binaries are auto-detected via .cmd shim paths and where.exe. Note:

• Ensure your agent CLI (claude, codex, openclaw, or pi-mono) is installed globally via npm
• OpenClaw Docker instance session scanning is not available on Windows (Docker Desktop uses WSL2 with different path layout)
• If using git, ensure core.autocrlf is configured correctly — .gitattributes enforces LF line endings

---

Full Changelog

See CHANGELOG.md for the complete list of changes.

Contributors

Thanks to everyone who contributed to this release!

v0.3.0

feat: dark/light mode toggle with CSS variable theming, theme toggle in NavBar, WCAG AA contrast for both themes

v0.2.3

fix: global install — add runtime deps to root package.json, resolve data dir to ~/.clawui

v0.2.2

fix: ship pre-built artifacts — fixes npm install -g failing with tsc: command not found

v0.2.1

Fix

• Fix npm install -g @clawui/cli failure — workspaces not supported for global packages (#5)

CI

• Automated npm publish via GitHub Releases with provenance attestation (#7)

v0.2.0 — Initial Public Release

ClawUI v0.2.0 — Initial Public Release

The Visual Orchestrator for Claude Code. Turn your linear CLI into a macro-planning Node Graph.

Key Features

• Blueprint DAG System — AI-powered task decomposition into dependency-aware directed acyclic graphs. Execute nodes in order with automatic context handoff via artifacts.
• Timeline Viewer — Rich session observability with collapsible tool pairs, markdown rendering, and inline continuation via AI-generated suggestion buttons.
• Session Management — Star, tag, filter, search, and archive across your entire Claude Code history. Project-scoped views with time-grouped lists.
• Context Boundary Control — Each blueprint node runs in an isolated session. Handoff artifacts carry completed-work summaries downstream, preventing context window explosion.
• Post-Completion Evaluation — AI evaluates completed nodes and can insert refinement or blocker nodes via graph mutations (INSERT_BETWEEN, ADD_SIBLING).
• Smart Recovery — Detects stale executions on server restart, recovers orphaned sessions, and re-enqueues interrupted nodes.
• Security Model — Localhost-only binding (127.0.0.1), rotating auth tokens, timing-safe comparison. Tailscale recommended for remote access.
• Four-Layer Data Architecture — Raw JSONL (read-only source of truth) → SQLite index → JSON enrichments → App state. Delete .clawui/ to reset; Layer 2 rebuilds in seconds.

Tech Stack

• Backend: Express + TypeScript + SQLite (better-sqlite3) + Claude CLI integration via expect
• Frontend: Next.js 14 + React 18 + Tailwind CSS 3 (dark theme)
• CI: GitHub Actions on Node.js 20 + 22 (typecheck, lint, build, 413 tests)

Getting Started

```bash
git clone https://github.com/ccchow/ClawUI.git
cd ClawUI && npm install && npm run dev
```

Open the secure URL printed in the terminal.