Loop Engineering Tools and Runtimes

Official Claude Code feature for keeping Claude working toward a measurable completion condition.

Durable coding goals, Multi-turn work, Clear validation conditions.

Requires a clear goal and stop condition.

Claude Code scheduled tasks & /loopOfficial

Official Claude Code scheduled prompt and /loop documentation.

Polling, Recurring prompts, Repeated checks.

Loops need hard stop rules and budget awareness.

Codex Follow a goalOfficial

Official Codex guidance for durable goals with validation loops.

Migrations, Refactors, Deploy retries, Long tasks with clear validation.

Vague goals can cause unnecessary changes.

Codex AutomationsOfficial

Official Codex automation feature for recurring tasks and background work.

Scheduled agent work, Recurring maintenance, Background checks.

Recurring loops should include budget limits, stop rules, and human approval gates.

Codex AGENTS.mdOfficial

Official Codex project instruction file guidance.

Durable project rules, Validation commands, Coding conventions.

Long or stale instruction files can confuse agents.

Codex SkillsOfficial

Official Codex skill documentation for reusable workflow packages.

Repeatable workflows, Specialized procedures, Reusable instructions.

Vague skill descriptions can trigger skills at the wrong time.

Codex SubagentsOfficial

Official Codex subagent documentation.

Parallel work, Specialized review, Multi-agent workflows.

Subagents need clear roles and coordination rules.

Claude Managed AgentsOfficial

Hosted Agents

Anthropic’s hosted long-horizon agent infrastructure.

Advanced long-running agent work.

High complexity; not recommended for a first, simple loop.

RalphifyExperimental

Loop Runtimes

Experimental runtime for loop engineering workflows using RALPH.md-style loop definitions.

Developers experimenting with local loop runtime patterns.

Treat RALPH.md export as experimental, not an industry standard.

continuous-claudeCommunity

Open-source orchestrator that runs Claude Code or Codex in a continuous loop — creating PRs, waiting for checks, and merging — with budget, time, and iteration caps.

PR babysitting and CI repair, Long multi-iteration coding loops, Parallel work via git worktrees.

Supports infinite loops (--max-runs 0) — always set --max-cost, --max-duration, or --max-runs.

claude-review-loopCommunity

Review Loops

Claude Code plugin that runs an automated code-review loop using Codex as an independent reviewer, with timestamped execution logs.

Maker-checker workflows, Cross-model review (Claude writes, Codex reviews).

Ships with --dangerously-bypass-approvals-and-sandbox as the default Codex flag — prefer the safer --sandbox workspace-write.

cobusgreyling/loop-engineeringCommunity

Open-source Collections · MIT

Practical loop engineering toolkit and patterns: a readiness audit, project starters, and a token-cost estimator, inspired by Addy Osmani and Boris Cherny.

Learning patterns, Auditing loop readiness, Estimating loop cost.

Use as inspiration; verify implementation details before production use.

agents-best-practicesCommunity

Harness / Safety · MIT

Provider-neutral Agent Skill for designing, auditing, and refactoring agent harnesses — installable for both Codex and Claude Code.

Harness design principles, Validation and authorization gates, Project instructions.

Adapt the rules to your own stack and permissions.

awesome-agent-orchestratorsReference

Open-source Collections

Curated list of tools for orchestrating AI coding agents — parallel runners, multi-agent swarms, and autonomous loop runners.

Discovering worktree tools, Multi-agent orchestrators, Coding agent UIs.

Directory quality varies; evaluate each project on its own merits.

awesome-agent-harnessReference

Open-source Collections

Curated, implementation-first list of ~300+ agent harness engineering resources, mostly GitHub projects.

Learning harness patterns, Safety and guardrail layers, Implementation ideas.

Reference list only, not a single product.

Mem0Community

Memory Layers

Useful memory-first framing for token-rich and token-poor agent loops.

Memory strategy, Long-running context, Token reduction.

Use conceptually unless directly integrating their product.

crystalCommunity

Desktop app to run multiple Codex and Claude Code sessions in parallel git worktrees. Deprecated in February 2026 in favor of its successor, Nimbalyst.

Parallel agent sessions, Worktree isolation, Comparing approaches side by side.

Deprecated as of February 2026 — the author directs new users to the successor, Nimbalyst.

Agent Orchestrators · Apache-2.0

vibe-kanbanCommunity

A Kanban board for orchestrating 10+ AI coding agents — each task gets a branch, a terminal, and a dev server.

Planning and assigning agent tasks, Reviewing diffs with inline comments, Coordinating many coding agents.

Community project — keep a human review gate before merge.

Agent Orchestrators · GPL-3.0

cmuxCommunity

Open-source platform for running multiple coding agents in parallel, with per-agent tabs, linked PR status, and an in-app scriptable browser.

Parallel agent execution, Tracking PR status per agent, Remote SSH workspaces.

GPL-3.0-or-later — check license compatibility before commercial use.

tuttiCommunity

Multi-agent orchestration CLI driven by a tutti.toml — chains prompt steps, shell commands, and agents into repeatable pipelines with git worktree isolation.

Config-driven multi-agent pipelines, Worktree isolation, Typed artifact hand-off between steps.

Community project — evaluate before production use.

parallel-codeCommunity

Desktop app to dispatch multiple AI coding agents in parallel, each in its own git worktree, with a built-in diff viewer and one-click merge.

Running several agents at once, Reviewing diffs before merge, Worktree isolation.

One-click merge still needs human review before merging.

Agent Orchestrators · Apache-2.0

sortieCommunity

Turns tracker tickets (GitHub Issues, Linear, Jira) into autonomous agent sessions. Agent-agnostic, shipped as a single Go binary with SQLite state.

Ticket-driven agent work, Issue-to-implementation across trackers, Self-hosted single-binary setup.

Autonomous ticket-to-PR flows need stop rules and human approval before merge.

NimbalystCommunity

Open-source visual workspace for building with Codex and Claude Code — parallel sessions on a kanban board with one-click git worktrees. The project crystal now directs users to.

Parallel Claude Code and Codex sessions, Kanban + worktree workflow, Inline diff review across editors.

Keep a human review gate before merge.