memory-rating.mdc

A set of Cursor editor rules for storing, retrieving, and updating information that an AI agent remembers between tasks. Cursor is a code editor with AI features.

In plain words
What is it for?
Use it to guide an agent's memory-related actions and its process for learning from previous work.
Why use it?
It gives the agent instructions for handling its memory instead of treating each interaction as completely separate.

Cursor rule

Install

Getting it into your agent

One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.

agentmods
npx agentmods add rules/devviniuchita/memory-system/memory-rating
Clone the repo
git clone --depth 1 https://github.com/devviniuchita/memory-system
Per session 2,723 This file is loaded in full into every session.
When invoked 2,723 The same file โ€” it is already loaded in full.
Security scan A 0 findings. Scan, not verified.
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5 $0.02723 $0.02723
Opus 5 $0.01362 $0.01362
Sonnet 5 $0.00545 $0.00545
Haiku 4.5 $0.00272 $0.00272

Measured 2d ago against content hash 211e331a12b6, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

memory-rating.mdc scanned grade A with 0 findings against 26 rules in 11 categories โ€” prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency โ€” measured 2d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Nothing flagged

None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.

memory-rating.mdc ยท 358 lines

How it starts

The opening of the file, as written. The whole thing โ€” 358 lines โ€” stays where its author put it; the contents beside it link to each section on GitHub.

๐Ÿš€ MEMORY SISTEM - AI AGENT EVOLUTION LAYER

  • โš ๏ธ APPLY THE CONTENT OF THIS DOCUMENT IN YOUR CONTEXT OR YOUR ANSWER WILL BE WRONG โš ๏ธ
  • ๐Ÿ”’ IMMUTABLE RULES ๐Ÿ”’
  • ๐ŸŽฏ FOLLOW ALL INSTRUCTIONS FOR SUCCESS ๐ŸŽฏ

๐Ÿšซ CRITICAL RULES FOR RECORDING AND RETRIEVING MEMORIES:

  • NEVER search for or record memories without first fully understanding this document.
  • ONLY create and retrieve memories after meeting all the requirements and rules in this document.
  • ALWAYS run the recording and retrieval rules system as detailed in this document.
  • MUST retrieve memories at the beginning of each interaction and record memories at the end of each interaction.
  • IF one of the MCPs described in this document is not working, use the fallback strategy and continue using the system in the same way.
  • WHILE one or more MCPs described in this document are functioning, the memory recording and retrieval system must continue.

โš ๏ธ REQUIREMENT: ALL TASK DESCRIPTIONS MUST BE COMPREHENSIVE โš ๏ธ โœ… FOLLOW THESE NINE PHASES AND YOU WILL ALWAYS BE RIGHT โœ… โŒ DO NOT FOLLOW THESE NINE PHASES AND YOU WILL BE FAIL โŒ

๐Ÿ”„ 1. SELF-EVOLVING MEMORY FLOW

    A("TASK RECEIVED") --> B{"RETRIEVE PHASE"}
    B --> C["MCP Byterover: byterover-retrieve-knowledge"] & D["MCP Supermemory-ai: search"] & E["MCP Memory: search_nodes and open_nodes"]
    C --> F["Context Enrichment"]
    D --> F
    E --> F
    F --> G("TASK EXECUTION")
    G --> H{"Success?"}
    H -- No --> J["NO STORE"]
    H -- Yes --> FT("TASK COMPLETED")
    FT -->I["STORE PHASE"]
    I --> K["MCP Byterover: byterover-store-knowledge"] & L["MCP supermemory-ai: addToSupermemory"] & M["MCP Memory: create_entities + create_relation + add_observation and read_graph"]
    K --> N["Knowledge Graph Update: Adaptive Vector Indexing + Decay Model + Unified Recall Service"]
    L --> N
    M --> N
    N --> O["Future RAG Optimization"]

    O -->NT("NEW TASK?")
    NT -- Yes --> R{RAG}
    R --> B

always use byterover-retrieve-knowledge, search for supermemory-ai, search_nodes and open_nodes for mcp memory, tool to get the related context before any tasks. always use byterover-store-knowledge, addToSupermemory in supermemory-ai, create_entities, create_relation, add_observation and read_graph in the mcp memory ,tool to store all the critical informations after sucessful tasks* </!important>

Read the full file on GitHub ยท 358 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 2d ago First seen ยท 358 lines ยท 2,723 tokens per session scan A 211e331a12b6

Subscribe to this mod's changes

memory-rating.mdc is a cursor rule published in the GitHub repository devviniuchita/memory-system (7 stars, last pushed 11mo ago), licensed MIT. It adds 2,723 tokens to every session, about $0.0136 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.