@cc-mem/core

The core of cc-mem: one memory for all your coding agents, kept in the auto memory Claude Code already writes. The agent plugins build on it:

| Agent | Package | |---|---| | pi | @cc-mem/pi | | OpenCode | @cc-mem/opencode | | DeepSeek Harness | @cc-mem/dsh |

Install one of those to use cc-mem in an agent. This package is for scripts and for writing a plugin for another agent.

Command line

Print a project's memory index the way an agent sees it, for scripts or hooks:

npx @cc-mem/core index --project <dir>

Library

• resolveLocation(env, project) resolves the memory directory as Claude Code does: the canonical git root, so worktrees share one memory; autoMemoryDirectory from user, local and managed settings; CLAUDE_CONFIG_DIR; the long-path hash. It also reports whether auto memory is turned off.
• memoryText(env, project, { writable }) builds what the plugins put in the prompt: where the notes are, how to treat them, and MEMORY.md within Claude Code's limits of 200 lines and 25,000 characters.
• writeMemory, editMemory and deleteMemory back the memory_write, memory_edit and memory_delete tools.
  • They check Claude Code's frontmatter format.
  • They never overwrite a file on create.
  • They delete one named memory file at a time, never MEMORY.md.
  • They record which agent wrote or changed a file.
• listDescription, writeDescription, editDescription, deleteDescription and fileNameDescription are the tool descriptions, shared so the tools read the same in every agent.

Use systemEnv() for the environment of the current process.

License

MIT. Not affiliated with or endorsed by Anthropic. "Claude" and "Claude Code" are trademarks of Anthropic.