Personal Beta · Control room

AgentRail control room for agent work.

AgentRail is for one operator running coding agents: it turns loose website-update chats into tracked workitems with checks, evidence, and a clean publish decision.

Open Install Center Open Release Center View Self-use SOP

Quickstart

Full install, downloads, checksums, and verification commands

Canonical host

www.agentrail.tech

Workitem railActive

IntakeGoal, scope, repo, and acceptance criteria.

ImplementCodex, Claude Code, or another worker updates the repo.

VerifySmoke tests, URL checks, artifacts, and review notes.

PublishMerge only when the evidence trail is clean.

state: IMPLEMENTING
next: run checks
target: www.agentrail.tech

Console screenshot or demo GIF placeholder

Workitem stateEvery task has a stage, owner, and next action.

Evidence trailChecks, changed files, and notes stay attached.

Release pathInstall script, release page, and downloads stay verifiable.

Personal-useBuilt for my own agentrail.tech operating loop.

How it works

STEP 01Create a workitem with the target page, repo path, scope, risk level, and acceptance checks.

STEP 02Run the worker through implementation, self-check, testing, review, retry, and blocked-state handling.

STEP 03Publish only after the artifact trail explains what changed, what passed, and what still needs judgment.

Primary paths

What is included

Installable Hermes skill package
Perpetual Engine runtime
CLI, tests, examples, and mock demo flow
Release assets, manuals, and product overview PDFs
Operator controls and recovery workflow

Problem it solves

Agent work is easy to start and hard to trust. AgentRail gives each task a visible state, a recovery path, and enough evidence that I can decide whether a website update is ready to merge.

About

AgentRail is maintained by BlueMarlin1999 as a personal beta workflow for agentrail.tech checks and updates.

Glossary

Hermes skill: the installable package that lets AgentRail plug into the Hermes operator environment.
Perpetual Engine: the local runtime and CLI that hold workitem state, checks, retries, and artifact records.
workitem: one tracked unit of work with a goal, scope, acceptance checks, state, and evidence.

Where it fits

Dispatch

Use Antigravity or a similar surface to shape the request, then let AgentRail hold the workitem, scope, acceptance criteria, and next action.

Workers

Send implementation or review work to Codex, Claude Code, or OpenCode while keeping the state machine and evidence outside the worker's chat window.

Operator control

When a task fails or needs judgment, the blocked state, retry count, changed files, and checks remain visible enough to resume or stop cleanly.

What this unlocks in practice

Parallel delivery

Route separate workstreams to different coding agents without losing stage control, audit trails, or operator visibility.

Best-tool routing

Use Claude Code where reasoning depth matters, Codex where repository automation matters, and OpenCode where provider flexibility matters.

Safer operations

When any worker fails, AgentRail preserves failure boundaries, blocked states, retry counts, and resume paths instead of pretending everything succeeded.

Current self-use status

Personal Beta

Self-use

Used as a private workflow rail for checking, updating, and recording AgentRail website tasks.

Verification Path

Mock + CLI

Use the install script, release package, CLI doctor, tests, and demo flow before treating a change as ready.

Commercial paused

Paused

No payment, resale, customer onboarding, or commercial support flow is active on this site right now.