CodingAgentAnalytics

Collect Codex, Claude Code, and Claude Cowork activity in one dashboard so the team can compare concurrent agent usage, spot blocked work, and improve how agents are assigned.

Get the hook installer Get the Claude plugin Read setup
Portable across macOS and Ubuntu/Linux. Build the archive and checksum with experiments-hooks-mgr:bundle.

What the team gets

Agent concurrency becomes visible.

The collector turns hook events into sessions, live working lanes, host leaderboards, peak concurrency, waiting states, rate limits, project hot spots, and Prometheus metrics.

PeakHighest number of overlapping agent lanes in the selected window.
LiveAgents working now, derived from recent work and in-flight tool spans.
WaitsPermission prompts, elicitation, notifications, and teammate idle events.
HostsMachine-level leaderboards using stable collector host identity.

Install paths

One collector, four agent surfaces.

Claude Cowork uses the plugin archive. The portable hook installer captures Claude Code, Codex, and Pi activity on macOS or Ubuntu/Linux without requiring a repository checkout.

Claude Cowork

Upload the plugin ZIP, configure the collector URL and secret, then start new Cowork tasks.

  1. Open Claude's Customize menu and choose Plugins.
  2. Add a personal/custom plugin file.
  3. Select coding-agent-analytics.zip.
  4. Set collector_url, collector_secret, and collector_agent=claude-cowork.

Claude Code

Download and extract the portable archive, then use the guided installer to preview and add the hooks.

tar -xzf coding-agent-collector-installer.tar.gz
cd coding-agent-collector-installer
bash install-collector.sh --target claude

Codex and Pi

Use the same archive for either target, or omit --target to install Claude Code, Codex, and Pi together.

bash install-collector.sh --target codex
bash install-collector.sh --target pi
bash install-collector.sh --remove

Operate it

Run local or shared analytics.

Use a loopback collector for personal experiments, or point the plugin at the shared collector when the team wants comparable host and agent metrics.

1Start the sink.
make run-collector SECRET=daybreak accepts hook events on :8788.
2Start the UI.
make run-ui serves sessions, dashboard, analytics, and /metrics.
3Verify one new session.
Open /sessions, then /analytics, and confirm your agent label and host appear.
4Use the numbers.
Compare peak concurrency, live agents, host leaderboard, waiting events, and rate-limit events during team experiments.

Local commands

Build and validate before sharing.

Build the portable hook installer

Build the macOS/Linux archive and checksum, then refresh this site's generated download copy.

node scripts/run-nx.js run experiments-hooks-mgr:bundle

Build the Claude plugin ZIP

Run this after editing plugin hooks, scripts, or skills. It writes dist/coding-agent-analytics.zip and refreshes this site's download copy.

cd tools/experiments/coding-agent-control-plane
python3 scripts/build_plugin.py

Team challenge

Raise concurrency without raising wait time.

The useful competition is not just who starts the most agents. Track peak concurrency, live agents, blocked projects, and average events per session together so higher throughput does not hide unclear prompts, permission churn, or rate-limit pressure.

Get the hook installer Get the Claude plugin Collector docs