If you have ever spent an afternoon debugging why your OpenClaw agent works on your laptop but fails on your CI server, HLock is the tool you have been waiting for. Released via Hacker News on October 29, 2026, HLock introduces harness.lock, a dependency tracking system designed to package the entire agent stackβ€”not just the model, but the MCP servers, skills, and configuration files that define its behavior.

The Lockfile for Agent Harnesses

HLock operates on a familiar philosophy for developers used to uv, cargo, or npm. The core command, hlock pin, records the current state of your harness, including specific versions of Claude Code, Codex, or autonomous agents like OpenClaw. This state is hashed and stored in harness.lock, ensuring that every dependency is pinned. When you move to a new machine, hlock sync rehydrates the exact setup, isolating it from any unlocked configuration on the host system.

Secrets and Team Collaboration

A critical feature for any production-grade tool is security. HLock automatically converts API keys and tokens into HLOCK_SECRET_* placeholders within the lockfile, allowing the configuration to be safely committed to Git. For teams, this solves the common issue of divergent agent setups. Contributors can commit their specific locks to .hlock/locks/harness..lock, enabling teammates to launch isolated environments using hlock launch --lock without overwriting their own local configurations.

Migration and Local Execution

The tool also addresses the fragmentation in the agent ecosystem. hlock migrate allows users to move a locked setup from one harness to another, such as switching from Claude Code to Codex, while preserving skills and MCP servers. Furthermore, hlock localize can convert any lock to run on open-source harnesses with local runtimes like Ollama or LM Studio, effectively cross-compiling your agent stack for offline or private execution.

Key Takeaways

  • HLock creates a harness.lock file to pin models, MCP servers, skills, and config.
  • API keys are replaced with placeholders to allow safe Git commits.
  • hlock migrate supports switching between harnesses like Claude Code and Codex.
  • hlock localize enables running agent stacks on local models via Ollama or LM Studio.
  • The tool integrates with an agent-driven meta-harness for automated setup and updates.

The Bottom Line

Agent reproducibility has been a messy, manual affair for too long. HLock finally treats agent configurations like proper software dependencies, which is exactly the infrastructure layer the OpenClaw ecosystem needs.

Sources

  • [Hacker News](https://docs.hlock.ai/)