Skip to main content

Getting Started with Claude Code

🆕 New page in this review

Everything on this page is new.

OpenBox integrates with Claude Code through its built-in hooks system, with no changes to how you use Claude Code, while every prompt, tool call, file edit, and shell command is governed, scored, and auditable.

This is a different door than the rest of Getting Started: the other integrations add governance to a runtime agent (a deployed workflow or graph). Claude Code isn't a runtime; it's the coding tool a developer uses. Here, OpenBox governs the dev session itself, the one writing and shipping the code that runtime agent will eventually run.

One Config Change​

Add OpenBox's hooks to your project's .claude/settings.json:

.claude/settings.json
{
"hooks": {
"UserPromptSubmit": [
{
"hooks": [{ "type": "command", "command": "npx openbox-claude-code hook user-prompt-submit" }]
}
],
"PreToolUse": [
{
"matcher": "*",
"hooks": [{ "type": "command", "command": "npx openbox-claude-code hook pre-tool-use" }]
}
],
"PostToolUse": [
{
"matcher": "*",
"hooks": [{ "type": "command", "command": "npx openbox-claude-code hook post-tool-use" }]
}
]
}
}

No changes to how you invoke Claude Code: start a session normally and the hooks run automatically.

Choose Your Path​

Claude Code 101​

Get the Claude Code concepts that matter for OpenBox (how a dev session maps to a governed agent) before wiring in governance.

I already use Claude Code​

Add governance to your existing project in a few minutes: install the hooks, start in observe mode, then turn on enforcement.

Show me the SDK reference​

Explore hook configuration, the event model, and troubleshooting.

What OpenBox Captures​

  • Every user prompt submitted to the session
  • Every tool call Claude Code makes (file reads/writes, shell commands, MCP tool calls) before and after execution
  • Governance decisions per tool call: ALLOW, CONSTRAIN, BLOCK, REQUIRE_APPROVAL, or HALT
  • Commits produced during the session, tagged with an OpenBox-Session trailer for lineage

What To Expect In The UI​

  • The dev session appears as a governed agent session, the same way a runtime agent session does
  • Tool calls show up as activities, same as any other integration
  • File writes and shell commands can be blocked, require approval, or proceed under a guardrail constraint, same as any governed operation
  • Session Replay works the same way for a dev session as for a runtime session

Next Steps​

  1. Read Claude Code 101 if you want the conceptual model first.
  2. Use Wrap an Existing Session if you already use Claude Code day-to-day.
  3. Continue to the Claude Code Developer Guide for hook configuration, event semantics, and troubleshooting.