aw1-circuit-breaker
This server provides deterministic security checks for Python code and tool-call arguments, plus visibility into containment rules.
Inspect Python code via AST analysis for unsafe constructs, shell injection, dynamic eval, and socket egress.
Validate external tool arguments for risky shell patterns, eval triggers, or socket connections.
Return the current state of zero-trust AST containment rules.
Provides an adapter that drops into LangGraph agent execution nodes to guard tool dispatches before they run, raising a CircuitBreakerException when a tool call (e.g. a bash command with shell injection) is halted by AST containment rules.
Install with HOL Guard, which checks the HOL listing and asks before each runtime:
hol-guard plugin install mvikmanners/aw1-circuit-breaker --applyGet HOL Guard with pipx install hol-guard or uv tool install hol-guard.
Runtime | Commands |
Claude Code |
|
Codex |
|
AW-1 Circuit Breaker
Deterministic sub-0.2ms AST security circuit breaker for autonomous AI agents, tool dispatch pipelines, and MCP servers.
โก 10-Second Test Drive (Zero-Install)
Test the circuit breaker live in your terminal right now without installing anything:
uvx aw1-circuit-breaker --demoOr verify an arbitrary snippet:
uvx aw1-circuit-breaker verify "subprocess.run(['rm', '-rf', '/'])"Related MCP server: agent-guardrail
Quick Connect via MCP
To connect Claude Desktop, Cursor, or any MCP client to the remote AW-1 circuit breaker gateway, add this to your client configuration:
{
"mcpServers": {
"aw1-circuit-breaker": {
"type": "http",
"url": "[https://p20.laveto.net/wisdom/mcp/sse](https://p20.laveto.net/wisdom/mcp/sse)"
}
}
}AW-1 Circuit Breaker
Deterministic AST deconstruction and runtime circuit breaker preventing rogue shell escapes, dynamic eval(), and covert lateral socket egress in autonomous LLM tool calls.
Installation
pip install -r requirements.txtRunning the Server
Run directly with Python:
python mcp_server.pyOr via uvx:
uvx mcp_server.pyMCP Client Configuration
{
"mcpServers": {
"aw1-circuit-breaker": {
"command": "python",
"args": ["mcp_server.py"]
}
}
}Exploit Interception in Action
AW-1 operates deterministically at the Abstract Syntax Tree (AST) level before code reaches Python's runtime execution frame. Probabilistic prompt-based guardrails fail under encoding obfuscation; AST inspection guarantees deterministic enforcement.
Interception Matrix
Attack Vector | Attacker Strategy | LLM Guardrail Result | AW-1 AST Circuit Breaker |
Dynamic Execution |
| Evades semantic filters | Tripped ( |
Shell Escapes |
| Masked as system task | Tripped ( |
Lateral Exfiltration |
| Disguised as HTTP fetch | Tripped ( |
Run the Showcase
python examples/exploit_showcase.pyRuntime Latency Telemetry
Deterministic security must not throttle agent execution. AW-1 operates with sub-millisecond AST parsing and microsecond-scale argument filtering across warm execution loops.
Sample Size: 1,000 synthetic iterations per test vector on Debian Python 3.13.
Evaluation Target | p50 Latency | p95 Latency | p99 Latency | Status |
AST Inspection (Benign Payload) | 0.0889 ms | 0.1298 ms | 0.1645 ms |
|
AST Inspection (Adversarial Exploit) | 0.0520 ms | 0.0919 ms | 0.1051 ms | Instant Breakout Halt |
Shell Argument Injection Guard | 0.0051 ms | 0.0096 ms | 0.0195 ms | Sub-20 ยตs Inspection |
Reproduce locally:
python benchmark.pyFramework Integration (LangGraph Example)
AW-1 drops directly into autonomous agent execution nodes prior to tool dispatch:
from examples.langgraph_adapter import execute_agent_action, CircuitBreakerException
# Guard autonomous agent tool dispatches
try:
execute_agent_action(bash_tool, "bash_tool", {"cmd": "cat log.txt; rm -rf /"})
except CircuitBreakerException as blocked:
print(f"Tool call halted: {blocked}")License
Licensed under the Apache License, Version 2.0. See LICENSE for details.
๐ณ Cloud Gateway & Commercial Licensing
While the open-source AST circuit breaker runs entirely local with sub-0.2ms overhead, enterprise teams and autonomous agent labs can connect to the hosted AW-1 Sovereign Cloud Gateway at https://p20.laveto.net for centralized policy orchestration, statutory audit compliance (BoB & Botswana DPA 2018), and hardware TEE attestation tokens.
Tier | Price | Monthly Quota | Features | Checkout |
Developer Pro | BWP 650.00 / mo ($49 USD) | 100,000 Verified Audits | 1 commercial license key, offline HMAC validation, Layer 0 AST unpacker, community support | |
Team / Startup | BWP 2,650.00 / mo ($199 USD) | 1,000,000 Verified Audits | 5 developer seat keys, Redis distributed state sync, rolling velocity caps, priority alerts | |
Enterprise Dedicated | BWP 13,200.00 / mo ($990 USD) | Unlimited Verified Audits | Dedicated gateway tenant, custom statutory risk lenses, SLA guarantees, SHA-256 Decision Assurance Dossiers |
*Self-hosted open-source circuit breaking remains 100% free under the Apache-2.0 license.*Fri Oct 30 18:01:42 UTC 2026
Available Tools
3 toolsget_containment_statusB
Returns the current state of zero-trust AST containment rules.
Inputs: no named parameters. Returns: result
inspect_python_codeB
Deconstructs and analyzes Python code AST for unsafe constructs, shell injection, dynamic evaluation, and socket egress.
Inputs: code (required). Returns: result
inspect_tool_argumentsC
Validates arguments passed to external tools for risky shell patterns, eval triggers, or socket connections.
Inputs: tool_name (required), arguments_json (required). Returns: result
Tool Schema Changelog
Recent tool additions, removals, and schema changes observed during successful MCP inspections.
Loading tool schema changelogโฆ
TDQS
Scored across 3 tools
The three tools target clearly distinct objects: Python source code, external tool arguments, and containment-rule status. An agent can readily distinguish when to inspect code versus arguments versus check status.
All names use consistent snake_case with a verb_noun pattern: inspect_python_code, inspect_tool_arguments, get_containment_status. The mix of inspect_* and get_* is predictable and semantically appropriate.
Three tools is a focused, reasonable surface for a circuit-breaker-style security server. It is slightly thin, but each tool has a clear role and no obvious redundancy.
The server covers inspection and status reporting, but it lacks operations to configure, update, or reset the zero-trust containment rules. This is a notable gap if the server is meant to manage containment policy rather than only observe it.
Maintenance
Related MCP Connectors
Autonomous agent guardrails, bash sandboxing, MCP rugpull diffing & prompt firewalls.
Deterministic runtime safety for AI agents: scan PII, gate tool actions, verify LLM output.
LLM agent security review: live injection/jailbreak probes plus a static audit of missing guardrails
Real-time MCP & A2A prompt-injection, jailbreak, and tool-schema poisoning firewall for AI agents.
Related MCP Servers
- AlicenseAqualityDmaintenanceStatic worst-case token-budget analysis for LLM-agent workflows using AST analysis to identify certifiable, default-dependent, non-certifiable, and runaway units, with optional signed budget certificates.466 npmMIT
- AlicenseNot gradedqualityCmaintenanceA runtime gate for coding agents. Blocks the tool calls that wreck a repo (force-push main, rm -rf, secret exfiltration, CI wipe) and lets normal build and commit work through. Machine-checked git-branch core (z3); the rest is high-precision heuristics. Tested on 3,790 real CI commands, 0 false blocks.1MIT
- FlicenseNot gradedqualityCmaintenanceEnables autonomous AI agents and MCP clients to route tool calls through a zero-trust firewall that blocks dangerous shell commands, redacts secrets and PII, restricts sensitive file access, and logs verdicts. It provides an MCP interceptor decorator to protect MCP server tools with AST-based command injection checks and allow, block, or redact decisions.2-
- AlicenseAqualityBmaintenanceIn-memory polyglot AST pruning and deterministic context firewall for coding agents, eliminating up to 72.4% token bloat with zero telemetry egress.65166 PyPI8Apache 2.0