graph-memory CLAUDE.md

Project instructions for using graph-based memory, which stores people, projects, decisions, and preferences as connected information.

In plain words
What is it for?
Use them when answering questions about prior conversations, people, projects, decisions, or preferences, and when maintaining the project’s graph memory.
Why use it?
They tell the agent when to look up past context and when it is appropriate to save or correct remembered information.

Instructions file

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 instructions/stevepridemore/graph-memory/claude-md
Clone the repo
git clone --depth 1 https://github.com/stevepridemore/graph-memory
Per session 632 This file is loaded in full into every session.
When invoked 632 The same file — 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 $0.00632 $0.00632
Opus 5 $0.00316 $0.00316
Sonnet 5 $0.00126 $0.00126
Haiku 4.5 $0.00063 $0.00063

Measured 2d ago against content hash 8c8f16470aa0, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

graph-memory CLAUDE.md scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — 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.

CLAUDE.md · 61 lines

How it starts

The opening of the file, as written. The whole thing — 61 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Graph Memory

You have access to a graph-based memory system via the graph-memory MCP server.

When to consult the graph

  • When the user asks about a person, project, past decision, or preference — call graph_query before answering from general knowledge
  • When the user references something from a previous conversation — the graph likely has context
  • When you're about to write a memory file — check graph_entities first to see if related knowledge already exists
  • When the user starts a new session — consider offering /graph-briefing if the project has graph data

How to use it

  • graph_query for structured lookups (entities, relationships, weights)
  • graph_cypher when you need a custom query (you know the schema — see /graph-ask skill)
  • graph_boost when the user confirms something you recalled from the graph
  • graph_weaken when the user corrects something the graph got wrong

Writing to the graph during conversation

You MAY write to the graph during conversation for high-confidence, explicit knowledge — but follow these rules:

When to write (call graph_relate)

  • User explicitly states a fact, preference, or decision ("I prefer X", "we decided Y", "Z works for me")
  • User introduces a person, project, or tool by name with clear context
  • User confirms or corrects something you recalled from the graph

When NOT to write (defer to the dream process)

  • Inferred context or things you're guessing from the conversation
  • Casual mentions without clear significance
  • Anything you're not confident about — the dream process will catch it later

Weight guidelines

Origin Weight
Explicit user statement ("I prefer X", "we use Y") 0.7
User confirmed recalled info ("yes, exactly") boost +0.15
Mentioned in context but not stated directly don't write — let dream handle at 0.3

Use specific relationship types — not just RELATED_TO

The full vocabulary (node types and edge verbs) lives in GRAPH_SCHEMA.md at the project root. Read that file before writing edges so you pick a specific verb (ABOUT, PART_OF, IMPLEMENTS, INSPIRED_BY, DEPENDS_ON, etc.) over generic RELATED_TO. RELATED_TO is the fallback only — use it when no specific verb fits.

Read the full file on GitHub · 61 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 · 61 lines · 632 tokens per session scan A 8c8f16470aa0

Subscribe to this mod's changes

graph-memory CLAUDE.md is an instructions file published in the GitHub repository stevepridemore/graph-memory (2 stars, last pushed 3mo ago), licensed MIT. It adds 632 tokens to every session, about $0.0032 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.