Cross-tool project memory and agent handoff. Use when setting up multi-agent workflows across OpenCode, Gemini CLI, Claude Code, and Codex.
This skill provides a file-based protocol for maintaining project context across different AI agents.
All project memory lives in Markdown files that any agent can read and edit:
AGENTS.md - Project identity, conventions, decisions.agent-context/current-work.md - What's being worked on now.agent-context/next-steps.md - What to do next.agent-context/handoff-message.md - For tool-to-tool transfersCopy the templates from templates/ and fill them in. See README.md for full instructions.
Run the scripts in your project root:
# Initialize
node init-universal-memory.js
# Capture state before stopping
node capture-state.js --task "..." --next-steps "..."
# Generate handoff package
node handoff.js --to [target] --task "..." --next-steps "..."
# Restore state when starting
node restore-state.js
AGENTS.md and .agent-context/current-work.mdAGENTS.md.agent-context/current-work.md and .agent-context/next-steps.md.agent-context/ for any existing handoffFor full script documentation and options, run:
node init-universal-memory.js --help
node capture-state.js --help
node handoff.js --help
node restore-state.js --help
This skill is file-based. No special tools or servers required.
Agents integrate by reading and writing the standard files. See README.md for detailed usage.