| name | agent-memory |
| description | Read or maintain agent-generated persistent memory when relevant context is folded or the user explicitly asks to remember stable information. |
Agent Memory
Use this skill only for stable, reusable agent knowledge:
- Use project scope for repository-specific facts and working conventions.
- Use global scope only for preferences or facts that apply across projects.
- Do not store credentials, secrets, transient task state, guesses, or conversation summaries.
- User-managed memory is controlled only through
/memory; never modify ~/.echo/memories.json.
- Never directly read, edit, patch, or delete files under
~/.echo/agent-memory/.
- Never change catalog or item enabled state. The user controls enablement through
/memory.
All agent memory operations must use scripts/memory.js from this skill directory through run_bash_command. Resolve the script from the absolute source path returned by use_skill; do not assume a global npm installation path.
Invoke the script with Node.js from the current project:
node '<skill-dir>/scripts/memory.js' <action> [flags]
Quote every value as one shell argument. For a single quote inside a single-quoted value, close the quote, append an escaped quote, then reopen it, for example: 'can'\''t'.
read --catalog <name> [--scope global|project]
add --catalog <name> --content <text> [--description <text>] [--scope global|project]
update-item --catalog <name> --item-id <id> --content <text> --scope global|project
update-catalog --catalog <name> [--name <new-name>] [--description <text>] --scope global|project
remove-item --catalog <name> --item-id <id> --scope global|project
remove-catalog --catalog <name> --scope global|project
validate
For add, omitted scope defaults to the current project; use --scope global only for cross-project knowledge. add requires --description when creating a catalog. read returns the resolved catalog scope and only enabled items. Use that scope and an item id from read for every update or removal. The script refuses to mutate disabled catalogs. validate checks accessible global/current-project catalogs without repairing files.
Successful commands print JSON to stdout. Invalid arguments, unavailable targets, disabled catalogs, or invalid storage print a diagnostic to stderr and exit non-zero. Never retry by editing storage files directly.
