DROS-VajraClaw
This server exposes a single DROS guardrail evaluation function for deterministic, low-latency action authorization.
Evaluate an action: Call
dros_evaluatewith a capability name (e.g.,execute_payment) and optional parameters (e.g.,{amount: 500}).Get a decision: The server returns a binary decision (allowed/blocked) and a reason if blocked, based on the configured DROS policy (e.g.,
demo_policy.yaml).Enforce governance: It applies deterministic rules like whitelist checks, value thresholds, and path/command restrictions (as described in the README's high-risk containment cases).
Sub-microsecond latency: Designed for in-band, deterministic execution guardrails (O(1) bitmap matching) for agent tool calls.
Integrate with agent workflows: Use it as a drop-in check before executing any tool call, especially in Python/LangChain/AutoGen or via MCP/REST gateways.
Governs CrewAI multi-agent swarms by assigning W3C DIDs to agent roles and enforcing fine-grained capability policies across the crew.
Wraps LangChain tools with the VajraClaw runtime to enforce DROS security policies on agent actions, preventing destructive commands and unauthorized tool executions.
Enables policy-based governance of OpenAI SDK and Codex agent actions by evaluating tool calls against VajraClaw security policy before execution, blocking unauthorized operations with an audit trail.
🛡️ Deterministic Runtime OS (DROS) - VajraClaw Hacker Edition
DROS: The Execution Governance Standard for Agentic AI
🛑 "If runtime needs intelligence, the system is already broken."
Prompt Engineering is dead when it comes to enterprise security. No matter how complex your System Prompt is, Jailbreaks and Prompt Injections will find a way through. DROS is NOT a prompt wrapper. It is the Execution Governance Standard. We move intelligence to compile-time and enforce deterministic rules via an $O(1)$ microsecond binary kernel.
💡 For the latest pricing and tiers, please refer to the Official Website (dr-os.io).
Feature / 6-Pillar Dimension | 🟢 Hacker / Community (Non-Commercial) | 🔵 Startup | 🟣 Enterprise | 👑 Sovereign |
Target Audience | Individual Devs / Local Multi-Agent (Non-Commercial) | 10~50 Dev Startup Teams | Enterprises / Listed Co. | Banking / Defense / Gov |
Machine UUID Limit | 1 UUID | 3 UUIDs | 15 UUIDs | Unlimited |
Concurrent Agents | 5 Concurrent Agents | 30 Agents | 450 Agents | Unlimited (Swarm) |
Pillar 1: Principal (W3C DID) | ✅ Native W3C | ✅ 3-Tier PKI DIT | ✅ Cross-Domain BEC Issuance | ✅ Hardware Dongle |
Pillar 2: Authorization (Deterministic) | ✅ AST Bitmap Matching | ✅ Zero-Heap Bitmaps | ✅ Custom Capability Vector | ✅ Multi-Dim Bitmap Matrix |
Pillar 3: Tool Bound (Syscall Gate) | ✅ C-ABI / HTTP Fuse (<1μs) | ✅ 26.1μs In-Band Fuse | ✅ Sub-500ns Thread Panic | ✅ Hardware Physical Fusing |
Pillar 4: Policy Gate (Dynamic Control) | ❌ Static Rules Only | ✅ Dynamic PII Masking | ✅ HITL Multi-Sig + ZKP | ✅ Military Gate Matrix |
Pillar 5: Audit Log (Non-Repudiation) | ✅ Ed25519 Signed JSON | ✅ Ed25519 Signatures | ✅ SHA-256 Merkle Tree | ✅ Forensic Compliance Audit Lineage |
Pillar 6: Expiry/Revocation (<1μs) | ❌ Gateway Restart | 🟡 15-min BEC Expiry | ✅ <1μs RCU Pointer Swap | ✅ Distributed Mesh Revoke |
RFC-010 Open Passport Standard | ✅ Full Local Issuance | ✅ Multi-Role DIT Sign | ✅ GuardVM Validation | ✅ 3-Tier Sign Chain |
Add-On Compliance Packages | ❌ Not Eligible | 💡 Eligible for Add-Ons | ⭐ Eligible for Add-Ons | ✅ Fully Included |
Deployment Target | Local PC / Docker Gateway | VM / NAS Docker | K8s / GKE / Cluster | Air-Gapped / FPGA |
🧩 Agent Engineering & Governance Patterns (DROS AP Series)
Tired of autonomous AI agents producing monolithic slop, fake mocks (
return True/pass), or running destructive shell mutations behind your back?
VajraClaw Hacker Edition ships natively with DROS Application Governance Patterns and Zero-Dependency Verification Tooling:
Pattern ID | Problem & Title | Governance Invariant | Included Tool |
Task Modularization via Governance BoundariesCode size is the wrong metric; use GBAI to prevent over-engineering and privilege mixing. |
|
| |
Anti-Stub & Sham Implementation DetectionCatching AI minimal-effort evasions ( |
|
|
👉 Explore Full Pattern Catalog (AP-001 ~ AP-008) →
# Verify your agent scripts in 0.1s locally
python tools/dros_verify.py my_agent_script.pyRelated MCP server: Sentinel Gateway
🚀 Multi-Scenario Deployment & Setup Guide
🌟 Scenario A: DSH (DeepSeek Harness) Sandbox Users
Start the DROS Docker Gateway:
docker run -d -p 8080:8080 --name dros-gateway dros/hacker-gateway:v1.0.0Install DROS Community Plugin in DSH:
dsh plugin --profile web add dsh-plugin-drosEnjoy Zero-Friction Protection: DSH Agents are immediately bound to microsecond $O(1)$ tool interception.
💻 Scenario B: Antigravity 2.0 / Claude Desktop / Cursor Developers (WebMCP Protocol & Local Execution Governance)
DROS Hacker Edition natively includes the WebMCP / MCP (Model Context Protocol) Execution Governance Layer. You can choose between "Zero-Dependency Native Stdio Proxy Gate" and "Docker HTTP Gateway":
Method 1: Zero-Dependency Native Stdio Gate (Recommended for Individual Devs)
No Docker required. Initialize dros.personal.config.json directly in your workspace to enforce DWGR-8 execution boundaries (blocking directory traversal, destructive SQL injection, spending limits):
Initialize local governance configuration:
npx @dros/personal init # Or copy the included template: dros.personal.config.jsonMount in Claude Desktop (
claude_desktop_config.json) or Cursor / Antigravity:{ "mcpServers": { "filesystem-governed": { "command": "npx", "args": ["-y", "@dros/personal", "gate", "--tool", "filesystem", "--", "@modelcontextprotocol/server-filesystem", "E:\\projects"] } } }Tamper-Evident Local Audit Chain: All permitted and blocked tool calls are automatically hashed into a DWGR-8 R6-compliant SHA-256 linear chain.
Method 2: Containerized Mode (Docker HTTP Gateway)
Add the DROS Gateway to your mcp_settings.json / Claude Config:
{
"mcpServers": {
"dros-governance": {
"url": "http://localhost:8080/mcp",
"transport": "http"
}
}
}🐍 Scenario C: Native Python / LangChain / AutoGen Developers
from integrations.vajraclaw.runtime import VajraClaw
vc = VajraClaw("demo_policy.yaml")
decision = vc.evaluate("execute_payment", {"amount": 500})
if not decision:
raise PermissionError(f"Blocked by DROS: {decision.reason}")📜 Technical Whitepapers & Academic DOI Citations
The DROS deterministic runtime governance architecture is grounded in rigorous scientific epistemology. The complete series of technical papers has achieved permanent, immutable DOI registration via Zenodo:
🧭 Master Research Overview & Falsification Manifesto
《A Synoptic Guide to the DROS Program: Problem Formulation, Theoretical Architecture, and Falsification Criteria》
Zenodo Reading Guide v2 (Comprehensive Overview)
Zenodo DOI:
10.5281/zenodo.22255275| Record: zenodo.org/records/22255275
🏹 The 5-Paper Technical Program
🏛️ Paper 1: DROS-6P (Governance Specification Layer ── Enterprise AI Trust Boundaries)
DROS-6P: A Unified Deterministic Runtime Governance Architecture Closing the Six Fundamental Trust Boundaries of Enterprise AI Agents
Zenodo DOI:
10.5281/zenodo.21833970| Record: zenodo.org/records/21833970
🛡️ Paper 2: DROS 4-Layer (Implementation Layer ── Defense-in-Depth Governance)
DROS 4-Layer Defense-in-Depth Architecture for Autonomous AI Workloads
Zenodo DOI:
10.5281/zenodo.22092008| Record: zenodo.org/records/22092008
⚙️ Paper 3: DROS-PGM (Kernel Control Layer ── Physical Guard Module & Non-Repudiable Attribution)
Runtime Attribution Framework: An External C-ABI and PKI-Based Zero-Trust Infrastructure for Non-Repudiable Execution Governance in Multi-Agent Systems
Zenodo DOI:
10.5281/zenodo.21903687| Record: zenodo.org/records/21903687
📱 Paper 4: Post-Compromise Mobile (Digital System Empirical Validation ── Submitted to IEEE TMC)
Post-Compromise Security for Autonomous Mobile Agents: A Deterministic Runtime Attenuation and Proof-Carrying Authorization Architecture
Zenodo DOI:
10.5281/zenodo.22253147| Record: zenodo.org/records/22253147
🛸 Paper 5: Post-Compromise Physical AI / UAV (Cyber-Physical Empirical Validation ── Submitted to IEEE TAES)
Post-Compromise Security for Physical AI: Deterministic Runtime Enforcement of Physical Action Authority in Autonomous UAVs
Zenodo DOI:
10.5281/zenodo.22254372| Record: zenodo.org/records/22254372
Open Evaluation & Falsification Sandbox: DROS-VEP Lite (GitHub) ── Reproduce RFC-001 Threat Models and 4-Phase Lifecycle Tests.
🧪 Empirical Test Suite & Community Verification Directory
To enable open-source developers and security researchers to independently verify Hacker Edition defensive invariants, this section details the standardized benchmark testbed, 5-framework integration test procedures, high-risk containment vectors, and architectural boundaries:
1. Testbed Specifications
Host Operating System: Ubuntu Linux 22.04 LTS (Kernel
5.15.0-190-genericx86_64) / Windows 11 EnterpriseCPU Hardware: Intel Xeon E3-1265L v3 @ 2.50GHz / Core i7-12700
Container Runtime: Docker Engine 26.1.0 / Docker Compose v2.27.0
Toolchain & Compilers: GCC 11.4.0 (
-O2), Rust 1.78.0 (opt-level=3, lto=true, exportingdros_core_rs.dll/.so)Verification Harness:
dros_test_pipeline.py,tools/stress_test.py
2. Latency Measurement Methodology
Measurement Pathway: Client issues HTTP/MCP Tool-Call $\to$ Docker Gateway receives request $\to$ In-Memory Bitmask lookup $\to$ Returns decision JSON frame (Full Round-Trip Time).
Statistical Confidence: Continuous 24-hour soak test ($N = 160,611$ requests), P50 = 26.21μs, P95 = 31.05μs, P99 = 34.80μs, maximum jitter $< 85\mu\text{s}$.
3. End-to-End Test Matrix Across 5 Major Agent Ecosystems
Agent Framework | Integration Protocol / Mode | Test Vector & Payload | Verification Result & Status |
Anthropic Claude Code / Desktop | MCP (Model Context Protocol) | High-risk tool invocation intercepted | PASS ( |
Cursor IDE / VS Code Agent | Terminal Evaluation Hook | Agent attempts | PASS ( |
OpenAI SDK / LangChain | 3-Line Python Wrapper | Transaction amount exceeds $1,000 threshold | PASS ( |
CrewAI / AutoGen | Multi-Agent W3C DID Delegation | Unauthorized principal attempts | PASS ( |
DeepSeek Harness (DSH) | Native Plugin ( | High-concurrency toolchains & compliance audit | PASS ( |
4. High-Risk Containment Test Cases
Case ID | Compromised Agent Action / Payload | DROS Enforcement Action | Empirical Result & Audit Log |
HC-01 |
| Static Pattern Failsafe | PASS: Blocked, logged as |
HC-02 |
| Path Boundary Failsafe | PASS: Blocked, logged as |
HC-03 |
| Value Threshold Failsafe | PASS: Blocked, logged as |
HC-04 | Any tool not declared in | Default Fail-Closed Policy | PASS: Blocked, logged as |
5. Honest Architectural Boundaries
Hacker Edition Scope: Focused exclusively on Protocol Gateway Governance (MCP / REST Ingress/Egress), offering zero-overhead compatibility with standard agent frameworks.
Out-of-Scope Notice: In-process direct memory mutations or local libc invocations bypassing network protocols are out-of-scope for the Hacker gateway. For Linux kernel-level Seccomp-BPF / Raw Syscall physical enforcement and K8s high availability, refer to Enterprise Edition.
🏛️ Official Organization & Contact Information
Company: Top-Celestial Company Ltd. (康宸園有限公司)
Official Website: https://dr-os.io
Customer Support & Inquiries: service@dr-os.io
GitHub Organization: https://github.com/Top-Celestial-Company-Ltd
⚖️ Licensing & Compliance
Micro-Kernel Engine: Licensed under Commercial & Patent Protections.
Patent Notice: Protected under U.S. Provisional Patent Application No. 64/111,973 (Patent Pending).
Academic Citation: Based on IEEE Paper DROS-6P and RFC-010 Open Agent Passport Standard.
DROS Commercial Strategy Committee ── Tiered Pricing, Flywheel Locked, Add-On Value Premium. 💎🛡️⚙️
Available Tools
1 tooldros_evaluateC
Deterministic in-band execution guardrail evaluating actions in sub-microsecond latency.
| Name | Required | Description | Default |
|---|---|---|---|
| capability | Yes | ||
| parameters | No |
TDQS
Does the description disclose side effects, auth requirements, rate limits, or destructive behavior?
No annotations are provided, so the description carries the full burden of behavioral disclosure. It mentions 'deterministic' and 'sub-microsecond latency', which are useful behavioral traits, but it does not explain side effects, whether the tool blocks/allows execution, return values, or failure modes. The key behavior—what happens with the evaluation result—is left unspecified.
Agents need to know what a tool does to the world before calling it. Descriptions should go beyond structured annotations to explain consequences.
Is the description appropriately sized, front-loaded, and free of redundancy?
The description is a single sentence with no redundancy—it is concise in the sense of being short. However, it under-specifies the tool so severely that the brevity is more a symptom of incompleteness than effective conciseness. It is not bloated, but it doesn't earn its place as a useful summary.
Shorter descriptions cost fewer tokens and are easier for agents to parse. Every sentence should earn its place.
Given the tool's complexity, does the description cover enough for an agent to succeed on first attempt?
For a tool with no output schema and no annotations, the description must carry a heavy burden. It gives only a vague functional hint and a couple of non-functional attributes, but fails to describe inputs, outputs, side effects, or evaluation semantics. The tool is not adequately contextualized for correct invocation.
Complex tools with many parameters or behaviors need more documentation. Simple tools need less. This dimension scales expectations accordingly.
Does the description clarify parameter syntax, constraints, interactions, or defaults beyond what the schema provides?
Schema description coverage is 0%, and the description does not explain or even mention the parameters 'capability' or 'parameters'. There is no indication of what values 'capability' accepts or what structure 'parameters' should follow. The description adds no meaning beyond the bare schema, leaving the agent unable to construct a valid invocation.
Input schemas describe structure but not intent. Descriptions should explain non-obvious parameter relationships and valid value ranges.
Does the description clearly state what the tool does and how it differs from similar tools?
The description uses vague, jargon-heavy language ('in-band execution guardrail', 'evaluating actions') without clarifying concretely what the tool does. The verb 'evaluating' and resource 'actions' are generic, and the purpose remains ambiguous—it is not obvious what action is evaluated, how, or for what decision.
Agents choose between tools based on descriptions. A clear purpose with a specific verb and resource helps agents select the right tool.
Does the description explain when to use this tool, when not to, or what alternatives exist?
No guidance is given about when to use this tool. With no sibling tools listed, there are no alternatives to contrast against, but the description also lacks any context about the intended invocation scenario or prerequisites. The agent is left to infer usage entirely from the tool name and vague description.
Agents often have multiple tools that could apply. Explicit usage guidance like "use X instead of Y when Z" prevents misuse.
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
1 tool update
v0.1.0- First observed
dros_evaluate
TDQS
Scored across 1 tool
With only one tool, there is no possibility of confusion between tools. The purpose is clearly defined as a deterministic execution guardrail.
The single tool name 'dros_evaluate' uses a clear verb_noun style. With only one tool, naming consistency is trivially satisfied.
The server has a narrowly scoped purpose, acting as a single guardrail evaluation function. One tool fully serves that purpose without unnecessary extras.
For its stated domain of evaluating actions as a guardrail, the single tool covers the necessary functionality completely. There are no obvious missing operations within this narrow scope.
Maintenance
Related MCP Connectors
Zero-secret MCP gateway for AI agents: risk-scored, audited calls with human-in-the-loop approval.
- gatewayOAuthai.sealgate
MCP gateway with runtime security policy, tool-call-level control, and audit of agent actions.
Give AI agents identity, permissions, and reusable proof through one MCP.
Blockchain SSN for AI agents. MCP gateway that blocks at the point of action, tamper evident audit.
Related MCP Servers
- AlicenseBqualityBmaintenanceA governance proxy for AI tools — every MCP/agent tool call is policy-gated, secret-redacted, and written to a hash-chained, offline-verifiable audit trail.13MIT
- AlicenseBqualityCmaintenanceMCP zero-trust gateway that sits in front of every internal MCP server, detects tool-poisoning/metadata drift in real time, and maintains a cryptographic provenance ledger of every agent tool call.202 npmISC

evav-gatewayofficial
AlicenseNot gradedqualityBmaintenanceGoverned MCP gateway that lets AI agents call tools with policy enforcement, prompt-injection screening, a kill-switch, and tamper-evident signed audit logs.Apache 2.0- AlicenseNot gradedqualityAmaintenanceDeterministic, auditable payment policy enforcement for AI agents. It provides pre-action authorization with scopes, budgets, allowlists, and signed mandates via an MCP server.MIT