SigAgent Architecture

SigAgent gives you two simple ways to see what Claude Code and Claude Agents are doing:

OpenTelemetry – for overall usage and cost tracking across your team
SigAgent hooks – for clear, step‑by‑step logs of every prompt, tool call, and result

OpenTelemetry

OpenTelemetry is great for company‑wide monitoring:

What you get: high‑level token usage, cost, and tool breakdown per session and project.
What you do not get: raw prompts, responses, or tool parameters/results. Message contents are not exported by Claude via OTel.
Included utility: a code‑change counter that estimates lines changed across edits (exported as an OTel metric).
Best for: organization‑wide visibility into Claude Code and Claude Agents usage and spend.
Agents parity: Claude Agents (CLI, Python SDK, TypeScript SDK) report via the same OTel interfaces as Claude Code.

Hooks allow full tracing of prompts and tool calls:

Installation: sigagent cli installs the sig-agent-marketplace, which includes the sig-agent-plugin.
Marketplace sandbox: see the repository for the CLI implementation and marketplace sandbox (sig-agent-marketplace repository).
Coverage: built‑in tools (e.g., ToDoWrite, Skills) and add‑on tools provided by MCP servers.
Captured data: full prompt text and model responses, tool parameters and results, plus timing and error data.
Upload path: hooks post traces to app.sigagent.ai; records are stored per‑organization in MongoDB and visualized in the UI.
Claude subscriber needs to use Pro or Max plan for hooks to be enabled.

Use the CLI to set org tokens and install the marketplace plugin:

Run npx -y @sigagent/cli setup and provide your organization access token.
Enable OpenTelemetry export and install the sig-agent-plugin from the marketplace prompts.
Restart Claude Code; verify traces and logs appear in the SigAgent dashboard.