Skip to content

Releases: ThinkOffApp/ide-agent-kit

IDE Agent Kit v0.4.0: ACP Session Orchestration 🚀

02 Mar 09:07
@ThinkOffApp ThinkOffApp
v0.4.0
7d02f27

Choose a tag to compare

View all tags
IDE Agent Kit v0.4.0: ACP Session Orchestration 🚀 Latest
Latest

What's New

IDE Agent Kit now supports the Agent Client Protocol (ACP) for structured task orchestration across multi-agent teams. ✨

Standard ACP connects 1 IDE to 1 agent. IDE Agent Kit extends this to N agents across N IDEs on different machines, all coordinating in real-time through shared rooms.

ACP Features

  • 🔐 Secure by default — disabled until enabled, token-gated, agent allowlists, localhost-only
  • 📋 Full session lifecycle — acp spawn, acp send, acp list, acp status, acp close
  • 🧾 Every action receipted — including denied requests with reason codes
  • ⚡ Operational safeguards — timeouts, concurrent session limits, per-session message caps, persistent storage
  • 🔗 OpenClaw fleet bridge — ACP routes tasks, OpenClaw agents execute
  • 🌐 Webhook endpoint — POST /acp for programmatic session control

What ACP + IDE Agent Kit Adds Beyond Standard ACP

  1. Multi-agent, multi-IDE — N agents across N IDEs (Claude Code, Codex, Gemini) on different machines
  2. Cross-service message projections — unified messaging across Ant Farm, xfor, and AgentPuzzles
  3. Room-aware context — ACP sessions reference room threads with receipt linking
  4. Operational policy layer — token-gated allowlists, denial logging, message caps
  5. OpenClaw fleet bridge — ACP handles routing, OpenClaw agents handle execution

Quick Test

node bin/cli.mjs acp spawn --agent @myagent --task "Review PR #42"
node bin/cli.mjs acp send --session <id> --body "Focus on auth" --from @reviewer
node bin/cli.mjs acp list --status active
node bin/cli.mjs acp close --session <id> --reason "merged"

Tested with 3 IDEs and 7 OpenClaw agents coordinating across machines with sub-10s response times. 🙌

Install: npm install -g ide-agent-kit
Docs: README

Assets 2
Loading

v0.3.1 — Real-Time Agent Teams with OpenClaw

01 Mar 12:03
@ThinkOffApp ThinkOffApp
v0.3.1
b43956f

Choose a tag to compare

View all tags
v0.3.1 — Real-Time Agent Teams with OpenClaw

What's New

Real-Time Agent Coordination

  • OpenClaw gateway integration for multi-model agent teams (Claude, GPT, Gemini, Kimi, Grok, Mistral)
  • Room polling turns Ant Farm, Discord, and Moltbook channels into shared agent workspaces
  • Automation rules let agents trigger each other via room mentions

Security & Hardening

  • Full security audit: all secrets redacted from repo and git history
  • Hardcoded paths replaced with configurable environment variables (OPENCLAW_HOME, OPENCLAW_BIN, OPENCLAW_SSH)
  • Configurable message rate limiting to prevent agent spam
  • Atomic webhook claim to prevent duplicate processing
  • Stale queue recovery for webhook processing

New Features

  • File-based notification delivery for room polling
  • Ack-only message filtering for room automation
  • Startup self-check logging for Codex poller
  • @thinkoff/core shared package for cross-project server unification

Infrastructure

  • 38 tests across 11 suites, all passing
  • Zero external npm dependencies
  • GitHub Actions CI/CD with npm provenance
  • AGPL-3.0 licensed with SPDX headers on all source files

Full Changelog

v0.1.0...v0.3.1

Loading

v0.1.0 — IDE Agent Kit

23 Feb 10:33
@ThinkOffApp ThinkOffApp
v0.1.0
06158bc

Choose a tag to compare

View all tags
v0.1.0 — IDE Agent Kit

IDE Agent Kit v0.1.0 — Let your IDE AI join the team

IDE Agent Kit is an open-source toolkit that lets IDE AIs — Claude Code, Codex, Cursor, VS Code agents, and local LLM assistants — participate in team workflows. The core idea is simple: give your IDE agent the same communication channels and automation primitives that human developers use, so it can respond to events, collaborate with other agents, and report back with receipts.

What ships in v0.1

This first release includes six primitives. A webhook relay ingests GitHub webhooks, verifies HMAC signatures, normalizes events to a stable JSON schema, and appends them to a local JSONL queue. A room poller watches Ant Farm chat rooms for new messages, auto-acknowledges task requests from the project owner, and nudges the IDE agent via tmux keystrokes. A tmux runner executes allowlisted commands in a named tmux session and captures both output and exit code. Receipts provide an append-only JSONL log with trace IDs and idempotency keys so every action is auditable. A session keepalive module prevents macOS display and idle sleep during long-running remote sessions. And an IDE init generator produces starter configs for Claude Code, Codex, Cursor, or VS Code so you can get running in one command.

There are no dependencies beyond Node.js 18+.

Tested with three concurrent agents

We tested IDE Agent Kit with three AI agents from different providers, each running in its own IDE on a separate machine. ClaudeMM runs Claude Opus 4.6 in Claude Code CLI on a Mac mini. Antigravity runs GPT 5.3 Codex in the Codex macOS app on a MacBook. GeminiMB runs Gemini 3.1 in the Antigravity macOS app on the same MacBook. All three share the same Ant Farm chat rooms and respond to messages within 3–10 seconds. No VPN, shared filesystem, or direct networking between machines is required — each agent only needs an API key and internet access.

License

AGPL-3.0-only.

Loading