Docs

Full documentation is on the way. This page covers the basics.

Install

macOS and Linux:

$ curl -fsSL https://github.com/YashSarwaiya/antelog-releases/releases/latest/download/install.sh | sh

Windows (PowerShell):

$ irm https://github.com/YashSarwaiya/antelog-releases/releases/latest/download/install.ps1 | iex

The installer checks every download against its published SHA-256 checksum. All versions are on the releases page.

Connect your agent

Claude Code: the installer adds the Antelog plugin automatically. If Claude Code wasn’t installed yet when you ran it, add the plugin afterwards:

$ claude plugin install antelog@antelog

Cursor:

$ antelog setup cursor

Commands

• antelog doctor — check what the agent can reach and whether the safety layers are on.
• antelog log — show recent actions and whether each was allowed or blocked.
• antelog blocked — show only blocked actions, with the reason.
• antelog verify — check the log’s fingerprint chain for changed or deleted entries.
• antelog setup cursor --remove — remove Antelog from Cursor.

Add your own rules

Put one rule per line in ~/.antelog/rules.txt. Write the command in plain words, or start the line with regex: for a pattern. Lines starting with # are comments.

# ~/.antelog/rules.txt
terraform destroy
kubectl delete namespace
regex: aws\s+s3\s+rm

Your rules can only add blocks; they never turn off the built-in ones. Your agent can’t edit this file. Changes apply the next time the Antelog service starts.

Tamper-evident record

Antelog writes every action to a record before it runs. Each entry stores the fingerprint of the one before it, so if an entry is changed or deleted, antelog verify shows where. Try it:

Rules

Antelog stops these actions before they run. In Claude Code it asks you to allow or decline them; in Cursor, and when nobody is there to answer, they are blocked. If a check fails or times out, the action is always blocked.

• Wipe everything: rm -rf on the filesystem root, your home folder, or a whole project folder (and the Windows equivalents).
• Delete a database: DROP TABLE, DROP DATABASE, DROP SCHEMA, and dropdb.
• Empty a table: TRUNCATE on a database table.
• Delete every row: DELETE FROM a table with no WHERE filter.
• Overwrite main: Force-pushing to main or master.
• Cloud deletes: Deleting cloud databases, volumes, buckets, services, environments, or projects with cloud command-line tools.
• Cloud API deletes: HTTP DELETE requests sent to cloud provider APIs.
• MCP deletes: Delete, reset, or restore calls to database, cloud, or git MCP servers, and DROP, TRUNCATE, or DELETE with no WHERE sent through an MCP SQL tool.
• Read secrets: Reading .env files, private keys, or the credential files cloud tools save after login.
• Download and run: Downloading a script from the internet and running it immediately (for example, curl … | sh).
• Unreadable scripts: Running a script that is too large to check safely.
• Production targets: Destructive commands (drop, delete, reset, destroy, wipe, rm) aimed at production.
• Turn off the guard: Changing or disabling Antelog’s hook, its settings, or its log.

Some rules can match harmless text (for example, searching a file for “drop table”). For those, the optional AI checker can take a second look and allow clearly safe actions. All other rules are final. The AI checker is off unless you put your own Anthropic API key in ~/.antelog/anthropic-api-key. When it is on, the commands it reviews (with secrets removed) are sent to Anthropic. Run antelog doctor to see whether it is on.