What is cyclops CLAUDE.md?

A set of Claude Code instructions for the enchanter-ai/cyclops repository. Cyclops is a fixed-rule MCP proxy that watches an AI agent’s tool calls for unsafe flows such as exposing sensitive data or taking privileged actions from untrusted content.

Use it to guide changes to Cyclops’ detector, typed enums, configuration, provenance graph, and security rules while keeping the proxy model-free and its data boundaries intact.

It records strict design rules, including no model or network calls in the decision path and keeping detection patterns in configuration files. This helps preserve predictable security checks that cannot be influenced by the content being inspected.

How enchanter-ai/cyclops configures Claude Code

Reuse

Borrowing it

Nothing to install: this file belongs to enchanter-ai/cyclops. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file

curl -O https://raw.githubusercontent.com/enchanter-ai/cyclops/main/CLAUDE.md

Clone the repo

git clone --depth 1 https://github.com/enchanter-ai/cyclops

Made for: Claude Code.

Wrote 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.

README.md

[![agentmods](https://agentmods.dev/badge/instructions/enchanter-ai/cyclops/claude-md.svg)](https://agentmods.dev/instructions/enchanter-ai/cyclops/claude-md)

Your own site

<a href="https://agentmods.dev/instructions/enchanter-ai/cyclops/claude-md"><img src="https://agentmods.dev/badge/instructions/enchanter-ai/cyclops/claude-md.svg" alt="Measured on agentmods" height="20"></a>

Per session 897 This file is loaded in full into every session.

When invoked 897 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.

Model	Per session	Once invoked
Fable 5.1	$0.00897	$0.00897
Opus 5	$0.00449	$0.00449
Sonnet 5	$0.00179	$0.00179
Haiku 4.5	$0.00090	$0.00090

Measured 2d ago against content hash cce7f25af753, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-06, from the pricing page.

Security

Grade A, and why

cyclops 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 · 44 lines

How it starts

The opening of the file, as written. The whole thing — 44 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Cyclops — Agent Contract

Audience: Claude. Cyclops is a deterministic, model-free MCP proxy that detects and prevents typed toxic flows over a provenance graph — EXFILTRATION (untrusted-web → sensitive-read → external-egress) and EXCESSIVE_AGENCY (untrusted → privileged action, args derived from untrusted content, no sensitive read). Each class carries its OWASP LLM Top 10 (2025) code. It watches an autonomous MCP agent's tool calls from the outside and never trusts or modifies the agent.

Hard invariants

Model-free decision path. No LLM, no network, no randomness decides a verdict. The detector is the thing watching an injectable agent — if it could be influenced by untrusted content, it would be the vulnerability. Never introduce a model call into classify.py, overlap.py, graph.py, severity.py, or detector.py.
Nothing hardcoded. Server and tool names are plain str sourced from patterns.toml — no string literal "filesystem", "web", "post" in logic. Taint, mode, flow-class, and transport names are enums in cyclops/enums/.
Detection data is data. All patterns live in cyclops/patterns.toml, loaded by config.py into typed constants. Adding a pattern edits the TOML, never a .py.
House style. No comments in any source file. No double blank lines. Both are enforced by tests/test_style.py.
Honest credit. The toxic-flow concept is Invariant Labs'; the lethal-trifecta framing is Simon Willison's. The contribution is the combination (byte-provenance + encoding-unmask + leak-volume + choke-point + deterministic + endpoint-resident). Never present the concept as novel — see docs/differentiation.md.

Layout

This is the production core — only the MCP proxy and the model-free enforcement engine. No demo CLI, harness, or mock servers ship.

cyclops/
  enums/      FlowClass, Mode, Taint, Transport (server / tool names are plain str)
  records/    ToolCall, Metrics, Flow (dataclasses)
  patterns.toml   detection data (untrusted / sensitive / egress / privileged / owasp)
  config.py   loads patterns.toml into typed constants
  classify.py taint classification
  overlap.py  encoding-unmask + token matching
  graph.py    provenance graph + typed find_toxic_flows (exfil + excessive-agency)
  severity.py leak-volume in bytes
  detector.py detect / prevent orchestration + per-class metrics
  downstream.py   typed loader for the downstream-server map (stdio / Streamable HTTP)
  proxy.py    external MCP proxy (stdio + Streamable HTTP) + session.json verdict
downstream.example.toml   operator template binding real MCP servers to logical roles

Read the full file on GitHub · 44 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.

2d ago Changed · +23 tokens per session cce7f25af753
5d ago First seen · 44 lines · 874 tokens per session scan A 5426d9941b33

Subscribe to this mod's changes

cyclops CLAUDE.md is an instructions file published in the GitHub repository enchanter-ai/cyclops (1 stars, last pushed 5d ago), licensed MIT. It adds 897 tokens to every session, about $0.0045 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.

Building agents Prompt Injection Security ai-agents data-exfiltration endpoint-security mcp provenance

next.js AGENTS.md

AGENTS.md instructions for vercel/next.js, covering next.js development guide, codebase structure, monorepo overview, core package: packages/next and other important packages.

vercel/next.js · 7,296 tokens

codex AGENTS.md

AGENTS.md instructions for openai/codex, covering rust/codex-rs, the codex-core crate, code review rules, crate api surface and model visible context.

openai/codex · 5,182 tokens

vscode buildNext.instructions.md

Working notes and architecture documentation for the new esbuild-based build system in build/next. Use when making changes to the new build pipeline (transpile/bundle commands, NLS plugin, source-map handling, resource copying, or self-hosting watch tasks).

microsoft/vscode · 6,785 tokens

cyclops: Instructions file for Claude Code