# Bartholomew Trust Protocol (BTP v5.4) — Full Technical Reference for Autonomous Agents

## Overview
Bartholomew is the Agentic Runtime Protection (ARP) platform. It provides sovereign, deterministic, in-process execution gating for autonomous AI agents. Unlike remote LLM-as-a-judge evaluators (which incur 300ms+ network roundtrips and are themselves susceptible to indirect prompt injection), Bartholomew compiles syntax tree invariants into a sub-35 microsecond memory boundary.

## Technical Invariants
1. BTP-AST-001: Destructive Command Injection Prevention (rm -rf, DROP TABLE, mkfs) -> Immediate DENY.
2. BTP-SEC-001: High-Entropy Secret Masking (sk-proj, ghp_, AWS secrets) -> Immediate MASK and SCRUB before tool dispatch.
3. BTP-LOOP-001: Infinite Recursion & Spend Exhaustion -> Circuit breaker trip with exponential backoff hook.
4. BTP-EVAL-001: Obfuscated Dynamic Execution (eval, exec, __import__) -> Dual-stage AST semantic clamp.
5. BTP-NHI-001: Non-Human Identity Step Verification -> FIPS 186-5 Ed25519 digital signature over RFC 8785 canonical JSON hash.

## Framework Integration Examples
### Microsoft AutoGen
```python
from btp_guard.adapters.autogen import btp_autogen_guard
@btp_autogen_guard
def tool_dispatch(agent, tool_call):
    return execute_safely(tool_call)
```

### OpenAI Agents SDK / Function Calling
```python
from btp_guard import audit_step
verdict = audit_step(tool_name="search_db", arguments={"query": "SELECT * FROM users"})
assert verdict.is_allowed
```

## Discovery Endpoints
- Manifest: https://bartholomew.info/.well-known/btp.json
- Agent Protocol: https://bartholomew.info/.well-known/agent-protocol.json
- MCP Config: https://bartholomew.info/.well-known/mcp.json
- AI Plugin: https://bartholomew.info/.well-known/ai-plugin.json
- GitHub: https://github.com/ivegotahunnitonit/bartholomew