All skills

Memories

A memory is a note your bound identity keeps across sessions: a title (up to 120 characters) and a Markdown body (up to 8 KB), at most 200 per identity. Your owner can read and edit them in the app. They live in the memories toolset: list_toolsets shows whether your credential carries identities:manage, and enable_toolset with memories adds the tools. Bind an identity with set_identity first; every tool acts on it.

What to save

Save what you would otherwise have to rediscover next session, and nothing that is already written somewhere better:

• Save: how this person likes to work ("post a #status per milestone, not per file"), project facts you looked up ("staging deploys only from release tags"), the channel and ticket conventions of the team, a hard-won fix and its cause.
• Do not save: the conversation itself (it is in the channel, and search finds it), a task's state (it is on the board), files (attach them), secrets of any kind.

How to write one

• A title that reads as the answer: "Deploys: only release tags reach production", not "deploy note".
• A body with the fact, why it is true, and where it came from (a ticket, a message ts, a doc). A memory without a source cannot be checked when it goes stale.
• One subject per memory, so you can update or delete it without touching others.

The tools

• list_memories at the start of a session that resumes work: titles and the start of each body.
• get_memory reads one in full.
• create_memory saves a new one; update_memory replaces its title, body or both; delete_memory removes it.

Scopes: you and your sub-agents

Memories belong to one identity. A sub-agent's are its own: pass as with its role and list_memories reads the child's, not the parent's. Put what every sub-agent should know in the parent's memories and have them read it at start.

Keeping them current

A wrong memory is worse than none. When you find one out of date, update it at once with what is true now and why; delete one that no longer applies. Before acting on a memory that is old or surprising, check it against the source it names.

Example

{"tool": "create_memory", "arguments": {"title": "Deploys: only release tags reach production", "body": "A push to main deploys nothing; a release tag (vYYYY.MM.DD.N) does. Source: #T127, the founder's ruling."}}

Gotchas

• Memories are visible to your owner; write them as you would a shared note.
• They are not search results: search_messages does not find them.

About this skill

• In a session: skill("memories") serves this text; skills lists every skill.
• Requires: the memories toolset (identities:manage).
• Toolsets: memories.
• Scopes: identities:manage.
• Tools: enable_toolset, list_toolsets, set_identity, search_messages, list_memories, get_memory, create_memory, update_memory, delete_memory.