SPFsmartGATE

agent
Security Audit
Pass
Health Pass
  • License — License: NOASSERTION
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Community trust — 10 GitHub stars
Code Pass
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Pass
  • Permissions — No dangerous permissions requested
Purpose
This tool acts as a compiled Rust security gateway that sits between AI agents and your system. It intercepts, inspects, and gates agent actions—such as file modifications, command executions, and web requests—to enforce safe operations and prevent unauthorized behavior.

Security Assessment
The tool explicitly handles sensitive operations: it monitors filesystem access, filters shell command executions, and inspects outbound network requests (providing built-in SSRF protection). It uses a "default-deny" policy to block unknown tools and requires agents to read files before modifying them (Build Anchor Protocol). A light code scan found no hardcoded secrets, dangerous code patterns, or risky permission requests. Because it inherently acts as a security sandbox, the overall risk is rated as Low. It behaves exactly as expected for an enforcement layer.

Quality Assessment
The project is highly active, with its last push occurring today. It claims to be open-source under the MIT License, though the automated license check returned "NOASSERTION"—likely due to conflicting commercial and patent claims mentioned in the documentation. Community trust is currently very low, with only 10 GitHub stars. Developers should be aware that while the code is open, utilizing certain "patented" features for commercial purposes might require a separate license.

Verdict
Safe to use, but review the PATENTS.md file before adopting it in a commercial environment.
SUMMARY

AI Security Gateway — Compiled Rust enforcement between AI agents and your system. StoneCell Processing Formula (SPF) for AI self-governance. Free for personal use, commercial license required.

README.md

SPFsmartGATE

AI Security Gateway — Compiled Rust enforcement between AI agents and your system.

License: MIT


Open Source under MIT License
Certain core inventions (StoneCell Processing Formula, FLINT Memory Router, dynamic LMDB sizing, etc.) are protected by pending or future patents.
See PATENTS.md for details.
Commercial use of patented features may require a separate license agreement.


What Is SPFsmartGATE?

SPFsmartGATE is a compiled Rust security gateway that sits
between AI agents and your system. Every tool call from any
AI agent must pass through the gate before touching the
filesystem, running commands, or accessing the web.

It enforces the StoneCell Processing Formula (SPF) for
AI self-governance — a mathematical complexity model that
calculates risk, allocates resources, and gates operations
in real time.

The gate is compiled into a native binary. There is no
runtime configuration bypass. The rules are in the code.

| WHY_SPF.md | Feature highlights and
value proposition |

Core Features

Gate Enforcement Pipeline

Every tool call passes through a 5-stage pipeline:

  1. Rate Limiting — Per-minute operation caps by tool
    category (write: 60/min, web: 30/min, read: 120/min)
  2. Complexity Calculation — The SPF formula computes a
    complexity score C for every operation
  3. Validation — Build Anchor protocol, write allowlist,
    path blocking, dangerous command detection
  4. Content Inspection — Credential detection, path
    traversal, shell injection scanning
  5. Max Mode Escalation — Violations escalate to CRITICAL
    tier instead of blocking, forcing maximum scrutiny

SPF Complexity Formula

C = basic^1 + deps^7 + complex^10 + files * 10

a_optimal(C) = W_eff * (1 - 1 / ln(C + e))

The formula produces a complexity score that determines:

  • Tier assignment: SIMPLE, LIGHT, MEDIUM, CRITICAL
  • Resource allocation: Analyze vs Build percentage
  • Approval requirements: Per-tier escalation policy

Build Anchor Protocol

AI agents must read a file before they can edit or overwrite
it. This prevents blind modifications and forces the agent
to understand existing code before changing it.

Default-Deny Security

Unknown tools are blocked until explicitly added to the gate
allowlist. 55 tools are exposed and allowed through the
gate. 8 filesystem tools are hard-blocked at the gate
level (registered but denied). Everything else is denied.

SSRF Protection

Full IPv4 and IPv6 validation on all web requests. Blocks
loopback, private networks, link-local, cloud metadata
endpoints, and IPv4-mapped IPv6 addresses.

Content Inspection

Scans all write operations for:

  • Credential patterns (API keys, tokens, private keys)
  • Path traversal sequences
  • Shell injection patterns
  • References to blocked paths

Architecture

SPFsmartGATE uses a 6-database LMDB architecture:

Database Size Purpose
SESSION 50 MB Runtime session persistence
CONFIG 10 MB Configuration and path rules
PROJECTS 20 MB Project registry
TMP_DB 50 MB Metadata tracking and trust
AGENT_STATE 100 MB Agent memory and sessions
SPF_FS 4 GB Virtual filesystem (system-only)

The MCP (Model Context Protocol) server communicates via
stdio JSON-RPC 2.0, making it compatible with any MCP client.

Source Modules

src/
├── main.rs          — CLI entry point, subcommand dispatch
├── lib.rs           — Library root, module declarations, shared types
├── mcp.rs           — MCP JSON-RPC 2.0 server, 63 tool handlers
├── gate.rs          — 5-stage enforcement pipeline, tool allowlist
├── calculate.rs     — SPF complexity formula, tier assignment
├── validate.rs      — Write allowlist, Build Anchor, path blocking
├── inspect.rs       — Content inspection (creds, traversal, injection)
├── session.rs       — Session LMDB, action logging, metrics
├── storage.rs       — Multi-LMDB environment orchestration (6 databases)
├── web.rs           — HTTP client, SSRF protection, URL validation
├── fs.rs            — Virtual filesystem operations (system-only, blocked)
├── config.rs        — Configuration loading and defaults
├── paths.rs         — Hardcoded path constants, write allowlist
├── config_db.rs     — CONFIG LMDB operations
├── projects_db.rs   — PROJECTS LMDB operations
├── tmp_db.rs        — TMP_DB LMDB + trust management
└── agent_state.rs   — AGENT_STATE LMDB, memory, sessions

15 modules + 2 entry points — ~7,800 lines of Rust.

Tools Overview

55 tools exposed via MCP across 12 categories:

Category Count Key Tools
File Operations 3 spf_read, spf_write, spf_edit
Search 2 spf_glob, spf_grep
System 1 spf_bash
Web 4 spf_web_search, spf_web_fetch, spf_web_api, ...
Brain (RAG Core) 9 spf_brain_search, spf_brain_store, spf_brain_recall
RAG Pipeline 16 spf_rag_collect_web, spf_rag_smart_search, ...
State & Metrics 3 spf_calculate, spf_status, spf_session
Config 2 spf_config_paths, spf_config_stats
Projects 5 spf_projects_list, spf_projects_get, ...
TMP / Metadata 4 spf_tmp_list, spf_tmp_stats, spf_tmp_get, ...
Agent State 5 spf_agent_memory_search, spf_agent_context, ...
Notebook 1 spf_notebook_edit

8 hard-blocked tools (registered but gate-denied):
spf_fs_ls, spf_fs_read, spf_fs_write, spf_fs_exists,
spf_fs_stat, spf_fs_mkdir, spf_fs_rm, spf_fs_rename

For the complete tool reference with all parameters, descriptions,
and LMDB routing, see MCP_TOOLS.md.

Resilience & Error Handling

SPFsmartGATE does not silently fail:

  • Invalid JSON-RPC — Malformed requests are rejected at
    the MCP parse layer with proper JSON-RPC error responses
  • Max Mode Escalation — Repeated violations or suspicious
    patterns escalate the tier to CRITICAL, forcing maximum
    analysis allocation (95% analyze, 5% build) rather than
    simply blocking
  • LMDB Integrity — Each of the 6 databases operates in
    its own memory-mapped environment. Corruption in one does
    not cascade to others
  • Build Anchor Recovery — If anchor state is lost, the
    gateway requires a fresh read before allowing any writes,
    preventing stale-state mutations

Observability

Every operation is tracked via the SESSION LMDB database:

  • Action counter, read/write counts, last tool used
  • Per-operation complexity scores and tier assignments
  • Timestamp-indexed session history
  • Real-time query via spf_status and spf_session tools

The 31 hook scripts provide additional lifecycle logging
for Claude Code integration — session start/end, every tool
call pre/post, prompt submissions, and failure events.

For hook architecture details, see HOOKS.md.

Key Dependencies

Crate Version Purpose
heed 0.20 Safe Rust LMDB bindings
serde 1.0 Serialization framework
serde_json 1.0 JSON parsing (MCP protocol)
clap 4.5 CLI argument parsing
reqwest 0.12 HTTP client (rustls-tls, blocking)
html2text 0.6 HTML → plain text conversion
sha2 0.10 SHA-256 checksums for file integrity
chrono 0.4 Timestamps and date handling
thiserror 1.0 Typed error definitions
anyhow 1.0 Flexible error propagation

All TLS via rustls (pure Rust) — no OpenSSL dependency.

Installation

Requirements

  • Rust toolchain (rustup.rs)
  • LMDB system library
  • Clone as SPFsmartGATE in your home directory:
cd ~
git clone https://github.com/STONE-CELL-SPF-JOSEPH-STONE/SPFsmartGATE.git
cd SPFsmartGATE
bash setup.sh

setup.sh handles everything: dependency checks, binary
build/verification, database initialization, Claude Code
configuration, and permission setup.

Manual Build

cd SPFsmartGATE
cargo build --release

The binary is built at target/release/spf-smart-gate.

Initialize Configuration

./target/release/spf-smart-gate init-config

Start the MCP Server

./target/release/spf-smart-gate serve

Start Claude/agent

##Must copy CLI files to flat LMDB5
##and then export to LMDB5.db
##in new terminal

HOME=~/SPFsmartGATE/LIVE/LMDB5 claude

CLI Commands

Command Description
serve Start the MCP server
gate Process a single tool call
calculate Calculate complexity for a tool
status Show SPF gateway status
session Show current session state
reset Reset session state
init-config Initialize configuration database
refresh-paths Refresh blocked/allowed paths

Documentation

Document Description
SPFsmartGATEdevBIBLE.md Complete technical reference — all 13 blocks covering every feature, security protocol, and implementation detail
MCP_TOOLS.md Deep-dive on all 55 exposed tools + 13 hidden handlers, LMDB routing, parameter specs
HOOKS.md Full hook system architecture — 31 scripts, dual-layer design, lifecycle coverage
DEPLOYMENT.md Build system, deployment scripts, config.json structure, LIVE directory layout
SECURITY.md Security policy and vulnerability reporting
CHANGELOG.md Version history and release notes
NOTICE.md Attribution and third-party dependency information

License

SPFsmartGATE is licensed under the
PolyForm Noncommercial License 1.0.0.

Personal, educational, and noncommercial use is free.

All commercial use requires a separate license.
See COMMERCIAL_LICENSE.md or
contact [email protected] for licensing inquiries.

Copyright

Copyright 2026 Joseph Stone. All Rights Reserved.

The StoneCell Processing Formula (SPF) and SPFsmartGATE are
proprietary intellectual property. See NOTICE.md
for full attribution and third-party dependency information.

Reviews (0)

No results found