Cortex CLAUDE.md

Development instructions for Cortex, a persistent-memory server for Claude Code that stores session information and builds domain profiles. They cover its Python implementation, storage options, testing, architecture, and releases.

In plain words
What is it for?
Use them when developing Cortex, choosing SQLite or PostgreSQL with vector search, running tests, checking installation behavior, or preparing releases.
Why use it?
They preserve useful context between sessions and give contributors a consistent way to build and verify changes to the memory system.

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/cdeust/cortex/claude-md
Clone the repo
git clone --depth 1 https://github.com/cdeust/Cortex
Per session 2,469 This file is loaded in full into every session.
When invoked 2,469 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 $0.02469 $0.02469
Opus 5 $0.01234 $0.01234
Sonnet 5 $0.00494 $0.00494
Haiku 4.5 $0.00247 $0.00247

Measured yesterday against content hash 72d5a225331a, method: parsed. Prices are Anthropic first-party input rates as of 2026-08-30, from the pricing page.

Security

Grade A, and why

Cortex 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 yesterday.

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 · 170 lines

How it starts

The opening of the file, as written. The whole thing — 170 lines — stays where its author put it; the contents beside it link to each section on GitHub.

@/.claude/rules/model-behavior.md @/.claude/rules/coding-standards.md

Cortex — Persistent Memory MCP Server

Persistent memory and cognitive profiling MCP server for Claude Code. Python 3.10+, FastMCP, Pydantic, numpy. Storage: SQLite by default (plugin installs, .mcpb/Cowork sandboxed launches) or PostgreSQL+pgvector as the opt-in upgrade (install-plugin.sh --postgres, CLI/dev mode, team databases) — see PRIVACY.md for the per-surface truth and mcp_server/infrastructure/backend_marker.py for how the plugin persists the choice.

Problem Statement

Claude Code sessions generate rich behavioral data (tool usage, session duration, first messages, keyword patterns) but this data is lost between sessions. Cortex mines this history to build a cognitive profile per domain and provides a thermodynamic memory system with heat/decay, predictive coding write gates, causal graphs, and intent-aware retrieval.

Build & Test

  • Install (dev): uv sync --no-default-groups --extra dev — SQLite backend adds --extra sqlite. Resolved from uv.lock, the same source CI installs from; resolving from the pyproject.toml ranges instead gives you versions CI never had (issue #253).
  • Environment preflight: python -m mcp_server.doctor (backend-aware check list, fix message per check)
  • Tests: pytest (full suite — current count: assets/badge-tests.svg, or run pytest --collect-only -q) · pytest tests_py/core/ (one layer) · pytest --cov=mcp_server --cov-report=term-missing- Lint BEFORE every commit: ruff check && ruff format --check — the CI enforces both; passing only ruff check is not enough.
  • Type gate (pyright, zero-diagnostic): resolve its environment from uv.lock (uv sync --no-default-groups --extra … --group typecheck), never from the pyproject.toml ranges — a range-resolved env reads a different type surface than CI and reports diagnostics CI never sees (issue #253). Exact command: CONTRIBUTING.md § Reproducing the pyright gate locally.
  • Release gate benchmarks (isolated, ephemeral container — the only source of truth for pre-tag/floor decisions): benchmarks/reproduce.sh. Do NOT gate a release against the live production database — same-day same-machine checks against it drift ±0.003 on LoCoMo MRR intra-day (measured 2026-07-14, benchmarks/results/repro/20260714-floors-rebaseline/) and nearly false-failed a floor that passes cleanly under reproduce.sh.
  • Iteration benchmarks: python3 benchmarks/{longmemeval,locomo,beam}/run_benchmark.py ⚠ Consolidation is OFF by default → scores collapse to ≈0%. This is a harness artifact (every candidate is prefiltered by the read-path heat gate before consolidation ever advances the stage), not a bug — pass --with-consolidation for representative numbers.

Read the full file on GitHub · 170 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. yesterday First seen · 170 lines · 2,469 tokens per session scan A 72d5a225331a

Subscribe to this mod's changes

Cortex CLAUDE.md is an instructions file published in the GitHub repository cdeust/Cortex (71 stars, last pushed 3d ago), licensed MIT. It adds 2,469 tokens to every session, about $0.0123 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 2026-08-30.