Skip to content

v1.4.0

Latest
Latest

Choose a tag to compare

View all tags
@pablodelucca pablodelucca released this 30 Jul 20:29
v1.4.0
5a950ff

Features

  • Claude Code Agent Teams visualization (#218) — Renders team leads and teammates as coordinated characters in both tmux and inline modes, with role labels, lead badges, token fuel gauges, tool activity, persistence, and coordinated cleanup. Builds on the early implementations in #79 and #177. Closes #65.
  • Claude Code sub-agents and teammates compatible up to v2.1.220 (#351) — Supports the newer Claude harness, where every Agent spawn runs in the background against an implicit per-session team. Named spawns become seated teammate characters with their own hook routing; unnamed spawns stay Subtask sub-agents whose transcripts are watched for live tool activity. Resumed leads re-latch to their newest team instead of stranding departed teammates or showing a phantom lead, and teammates take the free seat closest to their lead.
  • Context gauge on every agent (#351) — Shows each agent's context occupancy above its character, sized to the context window of the model that session actually runs.
  • Animated pets (#315) — Adds Claudio and Gitcat to the layout editor, autonomous walking and idle animations, petting interactions, saved placement, and custom pets loaded from external asset directories. Reimplements and supersedes #219 on the new architecture.
  • Carpets and workspace Areas (#316) — Adds three auto-tiling carpet styles with WYSIWYG color controls, paint/erase/eyedropper tools, undo, and persistence. Named Areas can be painted onto the office, mapped to workspace folders, and used to seat agents in the right part of the layout. Integrates the original carpet work from #213 and Areas work from #259.
  • Optional startup automation (#221) — Adds pixel-agents.autoShowPanel and pixel-agents.autoSpawnAgent settings to open the panel and launch an agent automatically when VS Code starts. Both remain off by default.

Standalone and Architecture

  • Layered architecture and provider-ready core (#236, #238, #273, merged through #275) — Separates shared contracts, the runtime server, the VS Code adapter, and the webview; centralizes lifecycle state in AgentRuntime and AgentStateStore; and removes runtime VS Code dependencies from the server. Provider, adapter, transport, and state interfaces now give future clients and agent providers stable integration points.
  • Standalone browser app and npm package (#273, #344) — npx pixel-agents starts a local Fastify server and browser office with WebSocket state delivery. The standalone UI now covers the core VS Code experience, including settings, diagnostics, asset reloads, layout import/export, Areas, carpets, and pets, while user state is centralized under ~/.pixel-agents/. Closes #120.
  • Multi-server hook discovery and fan-out (#344) — Lets the VS Code extension and one or more standalone servers run together. Each server registers independently, Claude hook events fan out to every live server, and each surface adopts only the sessions in its configured scope.
  • Published WebSocket protocol contract (#273) — Makes core/asyncapi.yaml the authoritative AsyncAPI contract and generates the shared TypeScript message bindings with CI drift detection.

Fixes

  • Agent and teammate lifecycle reliability (#287, #344) — Fixes idle/done state regressions, aligns teammate lifecycle behavior, and preserves agent tracking when terminals move between VS Code locations.
  • Standalone server and development reliability (#316, #344) — Honors an explicit --port, restores browser-mock hot reload, adds connection-state feedback, and fixes Windows e2e startup behavior.
  • Areas editor and seating polish (#316) — Correctly seats agents from custom-named workspace folders, fits four Area cards per row, and prevents the folder picker from being clipped.
  • File watching and session adoption (#352) — Fixes duplicate agent adoption on Windows, where case-sensitive path comparison let the same session be adopted twice, and closes a /clear race in which the external scanner could adopt a replacement transcript before SessionStart reassigned it.
  • Hook install failures and restored-agent rendering (#351) — Hook installation now reports failure instead of logging false success, and restored agents render regardless of the order in which layoutLoaded and existingAgents arrive. Closes #333, #334.
  • Standalone rendering of already-running agents (#349) — Sends layoutLoaded after existingAgents on webviewReady, so a standalone client that connects once agents already exist renders their characters instead of an empty office.
  • Workspace transcript discovery with hooks enabled (#330) — Keeps workspace JSONL discovery active when hooks are installed, so sessions are still found by file path rather than relying on hook delivery alone.
  • Cross-platform build and server tests (#289) — Fixes npm run build failing outright on Windows for paths containing spaces, and server tests that bypassed temp-home isolation and wrote to the real ~/.pixel-agents.

Testing and Release Infrastructure

  • Comprehensive Playwright e2e suite (#287, #344) — Expands coverage across VS Code and standalone, hooks-on and hooks-off lifecycles, Agent Teams, pets, carpets, Areas, and multi-server behavior. Adds a deterministic mock Claude process, narrated run videos, an in-repo coverage inventory, and combined Allure reporting.
  • Verified npm publishing pipeline (#344) — Adds package-contract tests, installed-tarball smoke verification, release tag/version checks, and provenance-ready npm publishing for the standalone package.
  • CI gating and platform coverage (#352) — Adds a single aggregate Required Checks job so branch protection can gate on one stable check instead of every shard name, and unblocks the e2e suite on macOS Tahoe by making video recording optional.

Maintenance

  • Move project links to pixel-agents-hq (#274) — Updates repository, issue, and community links after the GitHub organization migration.
  • Dependency and release-action updates (#224, #225, #226, #227)

Contributors

Thank you to the contributors who made this release possible:

  • @itsManeka — Animated pet system, bundled pet sprites, and pet interactions
  • @NNTin — Carpet system and foundational Playwright e2e infrastructure
  • @balgaly — Automatic panel display and agent startup settings
  • @elietwd — Cross-platform Windows build script and server test isolation
  • @snvtac — Workspace JSONL discovery with hooks enabled
  • @bezzborodth-tech — Standalone rendering fix for agents that already exist on connect
  • @srasantos — Reported the hook-install and restored-agent rendering bugs
  • @ErickGross-19, @ZenidX — Early Agent Teams implementations that informed the shipped design
  • @pablodelucca — Claude 2.1.220 sub-agent and teammate support, context gauge, Workspace Areas, layout-editor polish, and watchable narrated e2e review
  • @modtanoii — Architecture refactor collaboration, server and WebSocket design, and extensibility hardening
  • @florintimbuc — Architecture refactor, Agent Teams integration, standalone and npm package, multi-server support, file-watcher and CI hardening, e2e expansion, and release coordination

Community acknowledgements

Architecture and standalone direction

Earlier standalone implementations from @rollecode (#156), @MikaelDDavidd (#166), and @TimpiaAI (#63) demonstrated community demand and validated the browser SPA, WebSocket, and server discovery direction. These PRs were closed as superseded, with many thanks to their authors.

Independent fixes

Several community PRs independently diagnosed or fixed issues resolved in v1.4.0. Some overlapped with the final implementation, but each helped validate and strengthen the release:

Special thanks to @sakuramoon44, @Ralphive, and @joemanat1997 for contributing multiple fixes during this release cycle.

Assets 3
Loading