AgentRail

Repo-native harness for AI coding agents. Gives agents durable context, bounded issue execution, review loops, and verification gates — so agent work is easier to inspect, resume, and trust.

Works with Claude Code, Codex, Cursor, Hermes, or any custom command. You bring your own LLM keys.

Install

npm install -g @useagentrail/cli

Quick Start

Initialize AgentRail in your project:

cd your-project
agentrail init

Start with a grilling session to define your project context:

agentrail grill

Then create issues and run:

agentrail run issue 42 --agent claude

What It Does

• Context engine — builds scoped context packs from your repo so agents see what matters
• Bounded execution — plan → execute → verify phases with automatic retry on verification failure
• Review loops — PR review with finding triage (P0 → new issues, non-P0 → comments)
• AFK mode — unattended batch execution across issues using git worktrees
• Project memory — lessons, decisions, and failure patterns agents recall before acting
• Skills — curated workflow skills that guide agents through complex tasks

Commands

| Command | Description | |---------|-------------| | agentrail init | Install workflow files into a project | | agentrail grill | Challenge and refine a product idea | | agentrail run issue N | Execute a bounded plan/execute/verify loop | | agentrail run batch 1 2 3 | Run multiple issues in parallel worktrees | | agentrail afk | Unattended queue loop for labeled issues | | agentrail doctor | Check installation health | | agentrail status | Print workflow state | | agentrail context query "task" | Query the context engine | | agentrail memory recall "topic" | Recall project memory | | agentrail console | Dashboard status and setup | | agentrail upgrade | Upgrade managed files without overwriting edits |

Dashboard (Optional)

The CLI works fully without an API key. Set AGENTRAIL_API_KEY to unlock the dashboard:

• Agent operations console across repos and teams
• Run event streaming and cross-repo analytics
• Context pack audit trails
• Team collaboration and workspace sharing

export AGENTRAIL_API_KEY=your-key
agentrail console

Requirements

• Node.js 18+
• Python 3.9+ (for the context engine)
• gh CLI (optional, for GitHub integration)

License

Proprietary