herdr‑core: Persistent Terminals for Coding Agents

herdr‑core is a lightweight runtime that lets coding agents run in long‑lived terminals. The core itself draws no UI; it can be invoked from a shell, another agent, or a custom application.

Key Features

• Persistent Terminals – Terminals stay alive after you detach, and agents can spawn child agents that report back via `hc send` and `hc wait`.
• Cross‑Platform Builds – Static Linux builds for x86_64 and aarch64, macOS builds, and Windows x86_64 builds that require a POSIX shell (e.g., Git for Windows).
• Graceful Updates – `hc update` keeps the environment from the previous core; settings survive a restart for up to three minutes on Linux/macOS, longer on Windows.
• Background Execution – `hc run --bg` starts the machine’s default shell; `hc run --keep` preserves a terminal even if its program exits.
• Agent Skill System – `hc --skill` exposes an agent’s documentation, including delegation, waiting, reporting, and cleanup.
• Program Status Protocol – Agents can report status via OSC 7501; the core stores these reports.
• API and Events – `hc schema` outputs an OpenRPC schema; `hc events` and `hc get review` let you monitor changes.
• Security – Access control, state signing, and update verification are documented in the repository.

Practical Implications

Developers can run long‑running build or test agents that survive network interruptions. The ability to attach to any terminal means debugging can happen from any machine. The agent skill system encourages modular, reusable code.

How to Verify

Clone the repository, run `just install`, and start the core with `hc run`. Use `hc --help` for command references. The source code and CI configuration are publicly available on GitHub.

Further Reading

• herdr‑core GitHub Repository
• CONTRIBUTING.md
• License