Honeyrail

HoneyRail

Give it an engineering goal. Get back evidence.

HoneyRail is an open-source runtime for long-horizon, verifiable engineering work.

It orchestrates mature coding agents and deterministic tools through persisted DAG workflows, artifacts, evidence, evaluators, quality gates, human approvals, and reproducible engineering harnesses.

We don't build another coding agent. We harness the best ones.

HoneyRail starts from the Agent Gateway execution/control-plane subsystem: isolated git worktrees, tmux-backed agent sessions, task and worktree lifecycle state, project checks, evidence capture, REST/WebSocket/MCP automation, and an explicit human review step before merge. Run/Step orchestration, verification records, quality gate decisions, and the PostgreSQL Database Testing Harness Alpha now sit above that execution plane.

Why

Coding agents are becoming long-running engineering executors. Raw terminal sessions are useful, but they do not provide enough lifecycle management, isolation, verification, auditability, or approval control once agents start changing real repositories.

HoneyRail adds a local runtime around those agents so each engineering goal can become tracked work: a project, branch, worktree, session, checks, events, artifacts, and an explicit merge decision.

What It Does

• Runs persisted Run / Step DAG workflows above the atomic task/session/worktree lifecycle.
• Executes workflow steps through agent-task, shell, check, approval, and PostgreSQL harness executors.
• Creates isolated git worktrees per task and monitors project, task, session, worktree, and tmux state.
• Captures artifacts and evidence from checks, sessions, and harness steps.
• Runs deterministic evaluators plus async/custom evaluator registry entries.
• Persists QualityGateDecision records for pass, fail, operator override, and operator rejection.
• Supports human approval, rejection, commit, and merge review flows.
• Provides a PostgreSQL Database Testing