Shows
- Sessions and tool calls from the moment the computer is connected.
- How many tool calls succeeded, failed or were denied, and how long they ran.
Anthropic · Coding agent
Checked on against Anthropic’s documentation at code.claude.com (opens in a new tab). Tools change; this page says what was true on that day.
Free covers one computer with one coding agent. The team evaluation runs 14 days for up to 10 people. Neither needs a card.
Send Claude Code through the Governor gateway to see its models, token counts and estimated cost.
Prompts and responses; Code and file contents; Tool inputs and results; File, folder and repository names.
Each tool call is sent as a fingerprint of the tool’s name, not the name. Each project is sent as a random-looking id made on the computer from a private salt and the folder path.
Workspace hooks capture available activity and approved content. A separate supported terminal collector reports tokens, cache usage, latency, and estimated cost.
The current connector observes activity. Native tool denial and Enterprise inference hooks require a separately configured and verified enforcement adapter.
Source: code.claude.com (opens in a new tab), checked on 17 September 2026. Plan an enterprise rollout
curl -fsSL https://superamplify.com/connect/install.sh | sh -s -- --tool claude --ref governing_claude-codemacOS and Linux. It writes to ~/.superamplify, to Claude Code’s own settings and a background service for your user, and needs no administrator rights. Run it again with --remove to undo it.