Show HN: Stroq – a local audit of what your coding agent did and why

Stroq is an open-source, local security and auditing tool for AI coding agents, designed to show what an agent did, what content may have influenced it, and whether credentials appeared in its sessions. It scans existing Claude Code, Codex, and Cursor transcripts alongside local key files, reporting credential names and destinations without displaying key values. Its history review can assess sessions from before installation, tracing later commands to content the agent read; those verdicts apply today’s rules and do not prove what protections were active at the time. A local terminal card and browser console present findings and evidence without an account, telemetry, or sending data off the machine.

For future sessions, Stroq installs agent hooks that evaluate tool calls before execution. Policies can block risky patterns such as piping downloaded scripts into a shell or sending a known secret outward, and can ask the user to approve commands that repeat text the agent read. The console records decisions and their reasons, manages hooks and rules, and previews changes before applying them. In a test against 36 known attacks, the project reports 15 blocked, 21 requiring approval, and none passing through. Its coverage includes several coding agents, though the audit has limits: it identifies a secret as sent when the key is visible in the sending command, so scripts, files, or variables can carry credentials out without being detected.