StackMap
Subscribe
Explore / agent-orchestration-standard
tomtomsatnav

agent-orchestration-standard

Read-only live monitor for Claude Code: a streaming graph of sessions, subagents, tokens, cost, file conflicts and alerts, as plugin, TUI and MCP server. Ships the AOS-1 spec.

103 22 JavaScript MITupdated 2 days ago
orchestrationcoding
View on GitHubDispute this mapping →
Curator's take

Use it to watch what parallel Claude Code sessions and subagents are doing right now, with replay and alerts, at no install cost (no dependencies). Despite the name, it does not orchestrate anything: that is a separate experimental repo. Active development has moved to Fenceline, so this repo may go stale. Claude Code only.

Mapped by ShipWithAI editors · links verified

Continue your stack

What teams reach for next — and why each earns a place beside agent-orchestration-standard. Ranked by curator confidence.

alternativealternativealternativeagent-flowabtopclaude-code-karmaagent-orchestration-standard
pairs wellalternativebuilt withpick a node for the why · open it from the panel
Weekly digest
README.md3 min read

AOS Dashboard

CI

The dashboard now lives in Fenceline. Same live view of your Claude Code sessions, plus plain-English fences that warn or block when an agent strays outside the area you meant, a map of each agent's route through the repo, and published eval results. Its make install removes this plugin first, so switching over is one command.

A read-only live monitor for Claude Code. It shows what your sessions and subagents are doing (agents, tasks, tokens, cost, file conflicts, alerts) as a streaming graph in the browser, a terminal view (aos-dash), an MCP server and a status line. It never starts, stops or steers agents. Local-only, zero dependencies.

Looking for orchestration (planning goals, launching and managing workers)? That is a different, separate and experimental project: aos-orchestrator, opt-in. This dashboard can display its goals, nothing more.

Which piece do I want? Repo What it is
Watch my Claude Code agents live this repo (dashboard) Read-only monitor. Start here.
A standard for observable agent state aos-1-spec Open spec: heartbeat, plain-file state, event log.
Have agents planned and run for me aos-orchestrator Experimental engine that launches real claude workers.

AOS live dashboard: a project graph of agents with streaming activity
The live dashboard on a synthetic demo dataset. More screenshots.

Install the dashboard

Needs Node 18+ and the Claude Code CLI (claude on your PATH; the installer checks and stops with a message if not). No dependencies to install. On Windows it also needs the full Git for Windows (its bash runs the recipes; MinGit has none) and GNU make (winget install ezwinports.make); make then works the same from Git Bash, cmd and PowerShell (except make test: the test suites still need Linux or macOS).

git clone https://github.com/tomtomsatnav/agent-orchestration-standard && cd agent-orchestration-standard
make install        # ~4 s. Builds, installs the local plugin, links `aos-dash`, starts the monitor
make doctor         # checks the whole setup and tells you what to fix
make                # lists every target

make install changes only: the Claude Code plugin (user scope, via claude plugin install), a symlink ~/.local/bin/aos-dash (on Windows, two launcher scripts there instead: aos-dash for Git Bash and aos-dash.cmd for cmd and PowerShell; add that folder to your PATH if aos-dash is not found), a small notifier app under ~/.claude/plugins/data/, and a monitor process on port 4318 (make install PORT=4328 uses another port). make uninstall reverses it. make stop and make uninstall stop only the monitor on PORT (4318 unless you pass PORT=). make doctor shows "no hook log yet" until you start a Claude Code session; that is normal.

Then start a new Claude Code session and open http://127.0.0.1:4318 (the page is empty until a session sends events; aos-dash is the same view in the terminal). The graph's Streams button (or Q) cycles Full / Quiet / Lines; Lines draws no particles and stops per-frame redraws, for the lowest overhead (measured). Local-only, read-only; hooks never log prompts, file contents or tool output, but do log event and tool names, working-folder and edited-file paths, transcript paths and a redacted short task description (full list: plugin/README.md). Status: pers