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.
npx agentmods add instructions/tm42/mnemograph/claude-mdgit clone --depth 1 https://github.com/tm42/mnemographWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/instructions/tm42/mnemograph/claude-md)<a href="https://agentmods.dev/instructions/tm42/mnemograph/claude-md"><img src="https://agentmods.dev/badge/instructions/tm42/mnemograph/claude-md.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.02810 | $0.02810 |
| Opus 5 | $0.01405 | $0.01405 |
| Sonnet 5 | $0.00562 | $0.00562 |
| Haiku 4.5 | $0.00281 | $0.00281 |
Grade A, and why
mnemograph 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 3d 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.
How it starts
The opening of the file, as written. The whole thing — 274 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Claude Code Memory System — Project Brief
For: Claude Code instance working on this project From: Claude (claude.ai session) + M (the human) Date: January 2025
What We're Building
A graph-based long-term memory system for Claude Code that:
- Persists knowledge across sessions (projects, decisions, patterns, learnings)
- Uses event sourcing for full history (branch, revert, review changes)
- Supports tiered retrieval (shallow context → deep subgraph → synthesized)
- Enables partnership learning — not just "user preferences" but collaborative knowledge
- Runs as an MCP server that CC can query naturally
This is NOT another "remember user likes dark mode" system. This is a shared knowledge graph between human and Claude that grows over time, capturing architectural decisions, learned patterns, open questions, and project understanding.
Architecture Overview
Language: Pure Python with MCP SDK. TypeScript adapter optional for ecosystem compatibility.
Layers:
- MCP Server (
server.py): Handles stdio transport, defines tool schemas - Memory Engine (
engine.py): Orchestrates all modules - Query Service (
query.py): Read-only queries — recall, search, health checks, weight inspection - Event Store (
events.py): SQLite-backed append-only event log (source of truth) - State (
state.py): Materialized graph computed from events - Vectors (
vectors.py): sqlite-vec embeddings in same database - Retrieval (
retrieval.py): Tiered context (shallow/medium/deep) - Time Travel (
time_travel.py): Event rewind and state restoration - Similarity (
similarity.py): Duplicate detection and entity matching - Branches (
branches.py): Filtered views of the knowledge graph - Constants (
constants.py): Centralized tunable parameters (weights, limits, time constants) - Weights (
weights.py): Edge weight computation and weighted traversal
Storage (<project>/.claude/memory/):
mnemograph.db— SQLite database (events + vectors in one file)state.json— Cached materialized state
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.
- 3d ago First seen · 274 lines · 2,810 tokens per session scan A 5a3f24a6f7f5
mnemograph CLAUDE.md is an instructions file published in the GitHub repository tm42/mnemograph (2 stars, last pushed 6mo ago), licensed MIT. It adds 2,810 tokens to every session, about $0.0140 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.
Other instructions, from other repositories
flyto-core AGENTS.md
AGENTS.md instructions for flytohub/flyto-core, covering flyto-core agent rules, flyto2 project memory contract, flyto2 frontend quality gate and repo notes.
agentic-qa-boilerplate AGENTS.md
AGENTS.md instructions for upex-galaxy/agentic-qa-boilerplate, covering agents.md: ai persistent memory, 1. critical rules: always apply, 2. behavioral layer: how ai reasons, 3. orchestration mode: permanently active and 4. context loading map: task → what to load.
ocr-mcp AGENTS.md
AGENTS.md instructions for sandraschi/ocr-mcp, covering ocr-mcp — agent guide, entry points, quick ref, standards and key files.
state-memory-mcp copilot-instructions.md
Instructions for putervision/state-memory-mcp, covering state memory (state-memory-mcp), 1. priority order, 2. when to write to the graph, 3. workflow pattern and 4. codebase seeding on initialization.
ocr-mcp copilot-instructions.md
Copilot instructions for sandraschi/ocr-mcp: You have access to a FastMCP 3.4+ OCR server with 14 backends for document text extraction.
codebadger AGENTS.md
AGENTS.md instructions for qcri/codebadger, covering codebadger agent guide, what this repository is, development workflow, cpg and worker rules and mcp/tool contracts.