opencode_hermes

Open-source lab for autonomous AI agent experiments.

This repository is for comparing, adapting, and testing open-source autonomous agent ideas in the spirit of Hermes Agent, OpenClaw-style projects, and adjacent tool-using agent systems. The first priority is learning quickly while keeping experiments reproducible and easy to review.

Goals

• Track useful open-source autonomous agent patterns.
• Run small, reversible experiments before promoting anything to a larger implementation.
• Check agent context against the GitHub repository at the beginning of each project conversation or work session.
• Record hypotheses, outcomes, and follow-up tasks in a format future agents can reuse.

Context Refresh

GitHub is the source of truth for this project. At the start of a new conversation or work session in this project, run:

powershell -NoProfile -ExecutionPolicy Bypass -File scripts/update-context.ps1

The script fetches the latest repository state, fast-forwards the local branch when it is safe, and writes a local context snapshot to .agent-context/latest.md.

This is a session-start check, not a scheduled background refresh.

The .agent-context directory is intentionally ignored by Git. It is a local working memory cache for Codex, OpenCode, and other agents, not project history.

Experiment Flow

1. At session start, refresh context with scripts/update-context.ps1.
2. Create or update an experiment note under experiments/.
3. Keep changes small enough to inspect.
4. Record assumptions, commands, observations, and next steps.
5. Open issues for follow-up work that should survive beyond the current session.

Guides

• Hermes Agent beginner guide, Korean: detailed copy-paste setup and first-use guide for Windows, macOS, and Ubuntu.

Repository Layout

• AGENTS.md: project instructions for AI coding agents.
• docs/context.md: durable project context and decision log.
• experiments/: experiment plans, notes, and outcomes.
• scripts/update-context.ps1: local context refresh script.

License

MIT