A second look before your agent acts

TraceRook reviews what your coding agent is about to do before it does it. It runs inside Claude Code's PreToolUse hook, checks every proposed tool call against deterministic local rules, and sends ambiguous actions to TraceRook Cloud, where Anthropic Claude judges whether the action fits the task you asked for. Critical actions are blocked on the spot, high-risk actions wait for your decision, and every outcome comes with the evidence behind it.

Why TraceRook

• Review before execution. Dangerous tool calls are denied before the tool body runs. High-risk calls pause for a native review with a 45-second deadline.
• Local policy + Claude context. Deterministic rules catch credential exfiltration, destructive commands and remote execution on your Mac in milliseconds. TraceRook Cloud adds task-relevance, prompt-injection and intent-drift analysis with Claude Haiku 5.5.
• Private by design. Claude receives coarse task and action categories plus enumerated local signal codes. Raw commands, paths, code, file contents and transcripts never leave your Mac.
• Exact-action approvals. Allow once releases one specific pending invocation, bound to its fingerprint and nonce. Replays, stale notifications and late clicks are rejected.
• Native from top to bottom. A menu bar companion, session timelines, incident details and actionable notifications, built in SwiftUI.

Beta

TraceRook is in an invitation-only private beta for Apple Silicon Macs running macOS 26 or later. Invited members receive the download and a TraceRook Cloud access code. Billing isn't active during the beta; the planned price is $5/month, including TraceRook Cloud analysis. Request an invitation, then follow the installation guide.

Agent	Status
Claude Code	Supported — pre-execution blocking, Cloud analysis and native review
OpenAI Codex	Coming soon

About

TraceRook is built by its co-founders:

• Kyle LePrevost — [email protected] · hardcidr.com
• John Yang — [email protected] · LinkedIn

Repository layout

Path	Contents
App/	SwiftUI app, menu bar companion and review window
Agent/	Per-user background service
HookCLI/	tracerook-hook, the bridge invoked by agent hooks
Packages/	Shared Swift modules: contracts, adapters, privacy, rules, core, fixtures
Tests/	Swift Testing suites
cloud/	TraceRook Cloud — Cloudflare Worker serving api.tracerook.dev
backend/	Python local API for development and demos
website/	Static site and documentation for tracerook.dev

Build from source

./scripts/build.sh
open build/TraceRook.app

open build/TraceRook.app --args --demo launches with bundled sample sessions. See the development guide for Xcode, tests, the Cloud Worker and the website.

Documentation

Guide	Contents
Product documentation	Setup, policy, approvals, Claude analysis, privacy and integrations
Installation	Download, open, connect TraceRook Cloud and configure Claude Code
Development	Build, run and test every component
Architecture	Processes, modules and transports
Privacy	What stays on your Mac, what reaches Claude and what is retained
Threat model	Trust boundaries and how each is enforced
Agent compatibility	Hook formats and decision output
Cloud API	The /v1 protocol served by TraceRook Cloud
Security	Security model and vulnerability reporting

License

MIT · Copyright © 2026 Kyle LePrevost.

Claude is a product of Anthropic. TraceRook is an independent project; no affiliation or endorsement is implied.