memory-crud

name	memory-crud
description	Memory CRUD operations — add, get, update, delete memories, and batch operations. This skill should be used when the user asks to create, read, update, or delete memories, or perform bulk memory management.

Memory CRUD Operations

Covers all memory lifecycle operations — adding, retrieving, updating, deleting, and batch management.

Available Tools

Tool	Purpose	Required Params
add_memories	Add new memories	messages[]
get_memory	Get single memory	memory_id
get_all_memories	List all memories	—
update_memory	Update content	memory_id, data
delete_memory	Delete memory	memory_id
batch_update_memories	Bulk update (max 100)	memories[]
batch_delete_memories	Bulk delete (max 100)	memory_ids[]

Adding Memories

Tool: add_memories

Messages Format

add_memories({
  messages: [
    {"role": "user", "content": "The project uses React 18 with TypeScript"},
    {"role": "assistant", "content": "Noted: React 18 + TypeScript stack"}
  ],
  user_id: "project-alpha",
  metadata: {"category": "tech-stack", "project": "alpha"},
  infer: true
})

Parameters

Param	Type	Required	Description
messages	array	Yes	Array of {role, content} objects
user_id	string	No	Scope to a specific user
agent_id	string	No	Scope to a specific agent
app_id	string	No	Scope to a specific app
run_id	string	No	Scope to a specific run
metadata	object	No	Key-value tags for categorization
infer	boolean	No	true = extract facts from conversation, false = store literally

When to Use infer

infer	Use Case
true (default)	Conversational input — system extracts key facts automatically
false	Structured facts — store the exact text as-is

Getting Memories

Single Memory

get_memory({ memory_id: "abc-123" })

All Memories with Pagination

get_all_memories({
  user_id: "project-alpha",
  page: 1,
  page_size: 50,
  limit: 100
})

Pagination Parameters

Param	Type	Default	Description
limit	int	0 (all)	Maximum total results
page	int	0	Page number
page_size	int	0	Results per page

Updating Memories

Tool: update_memory

update_memory({
  memory_id: "abc-123",
  data: "The project migrated from React 18 to React 19"
})

Important: The data parameter must be a string, not an object.

Deleting Memories

Tool: delete_memory

delete_memory({ memory_id: "abc-123" })

Best practice: Always get_memory first to show what will be deleted, then confirm with the user.

Batch Operations

Batch Update (max 100)

batch_update_memories({
  memories: [
    { "memory_id": "abc-123", "data": "Updated fact 1" },
    { "memory_id": "def-456", "data": "Updated fact 2" }
  ]
})

Batch Delete (max 100)

batch_delete_memories({
  memory_ids: ["abc-123", "def-456", "ghi-789"]
})

If more than 100 items: Split into chunks of 100 and execute sequentially.

Common Workflows

Add and Verify

1. add_memories(messages, user_id) -> get created IDs
2. search_memories(query) -> confirm stored correctly

Safe Update

1. get_memory(memory_id) -> show current content
2. Confirm change with user
3. update_memory(memory_id, new_data)
4. get_memory(memory_id) -> verify update

Cleanup Stale Memories

1. get_all_memories(user_id) -> list all
2. Identify outdated entries
3. batch_delete_memories(stale_ids) -> remove
4. get_all_memories(user_id) -> verify cleanup

Bulk Content Correction

1. search_memories(query) -> find memories to fix
2. Prepare updates array
3. batch_update_memories(updates) -> apply all at once

Common Errors

Error	Cause	Fix
Memory not found	Invalid memory_id	Verify ID with search_memories first
Batch too large	>100 items	Split into chunks of 100
Empty results	Missing user_id scope	Add user_id parameter
data must be string	Passing object to update	Convert to string
Duplicate content	Not checking before add	search_memories before add_memories

Best Practices

1. Scope with IDs — always use user_id/agent_id for multi-tenant setups
2. infer=true for conversations — let the system extract key facts
3. infer=false for structured data — store exact text when precision matters
4. Search before adding — avoid duplicates
5. Batch for efficiency — use batch operations instead of loops (max 100)
6. Verify destructive ops — get_memory before delete, confirm with user
7. Use metadata — tag memories for easier categorization and filtering