知Yizhi Daily
← Back to this edition

Conversations

Stroq launches local AI coding agent audit tool to trace command origins and key exfiltration

Edition 2026-11-06 · 08:00 Beijing time

The Show HN project Stroq can read local sessions from Claude Code, Codex, and similar tools, auditing how many commands came from text the agent had read and which key names have left the machine. By default, the tool runs only locally, creates no account, and has no telemetry; it outputs key names without displaying their values.

The Stroq command line can read sessions from the last 30 days, key files, and folders where agents have worked, and writes a small cache to the .stroq directory in the user’s home directory. The project description says it does not send data and does not require an account; reports print only key names, not key values. It requires Node.js 22 or later and uses the Apache-2.0 license.

In addition to auditing historical sessions, Stroq can use hooks to make judgments before tool calls execute: it rejects curl | sh and blocks outbound requests carrying known keys; when a command repeats text that an agent had read, it first asks for user consent and points to the relevant text. The console runs on 127.0.0.1 and shows which agents are actually connected to Stroq and the reason for each decision.

The project provides commands including replay, sent, init, why, and attack. Among them, replay can replay the last 30 days of Claude Code sessions without installing hooks; attack is used to watch Stroq block 36 known attacks in a sandbox; init connects hooks into subsequent agent sessions. The materials say historical checks do not execute session contents and do not send information externally.

Original sources

Hacker News ↗
Back to this editionOpen interactive reader