Using VeriCommand with Claude Code
VeriCommand runs alongside Claude Code on Windows. You keep Claude Code and your own plan; VeriCommand adds a guard that checks each command before it runs, and one hash-chained record of what the session did.
Before a command runs
The guard is a Claude Code hook that runs before each command (a PreToolUse hook). It asks before destructive commands such as git reset --hard, and refuses root and home deletes, secret-file reads such as cat .env, and file edits into the board’s own record. It records the ask or the refusal, not your answer.
When it can’t read a command (a command held in a variable, an encoded body it can’t decode, a line over 32 KB) or can’t finish checking it within 8 seconds, it asks you instead of letting it through. If the guard hook can’t start or doesn’t answer in time, Claude Code handles the command under its normal permission rules.
It reads commands, not intent. Known gaps: aliases, script files, shell functions, indirect and symlinked paths, secrets already in the environment, and some compound PowerShell and quoting edge cases. A seatbelt, not a sandbox. What the guard covers
A record of the work
Claude Code joins the board through VeriCommand’s MCP server. Tasks, dispatches, progress signals and returns land on one hash-chained record on your machine, re-checked every time it is read. A return binds to the dispatch that authorized it, and findings an agent grades itself on count as assertions, not clearance.
The record proves integrity, not correctness. It shows what was recorded; it does not make the work right. How the audit trail works
After compaction or a new session
When Claude Code compacts a conversation, or a session ends and a new one picks up the work, the next stretch starts from a summary or from scratch. With the record, it can read the recorded state of the work instead of relying on that summary alone or re-reading every file it touched. Resuming a session
A second opinion from another vendor
An independent check sends Claude Code’s result, through NorthGate’s server, to a model from a company other than Anthropic. That model checks it against the decisions you recorded and quotes both sides. You keep the final decision.
A check sends the selected decisions and the result being checked, plus which AI company your agent runs on (so a different one checks it) and whether it ran by hand or automatically. Free includes 3 checks a month, run when you click; Pro ($39/month) includes 200, and up to 100 of them can run automatically after you say yes once. Every network path
Setting it up
- Install the signed Windows app (Windows 10 and 11).
- Follow Welcome and Connect in the app. They say why each step is needed and name every file Connect writes.
- Other agents can join the same board and record, free: Codex, Cursor, Kimi and other MCP-capable tools. The guard covers Claude Code sessions only.
Prefer the command line? The MCP server is also published on PyPI as vericommand. For developers
Claude and Claude Code are made by Anthropic. VeriCommand is made by NorthGate Strategic LLC; it is not an Anthropic product and is not endorsed by Anthropic. This page replaces an earlier first-person essay dated 2026-09-01.