t81-vm-runner

name	t81_vm_runner
description	Execute T81 code with deterministic guarantees and policy enforcement
metadata	{"openclaw":{"os":["darwin","linux"],"requires":{"bins":["t81"]}}}

T81VM Runner Skill

This skill executes T81Lang programs through the T81VM with deterministic guarantees, policy enforcement, and complete audit trails.

Usage

When the user needs to run T81 code with guaranteed determinism and governance, use this skill to execute programs through the T81VM.

Commands

Run a T81 program:

run <program_file> [--policy <policy_file>] [--weights-model <model_file>] [--trace] [--json]

Run with debug information:

debug <program_file> [--breakpoints <addr1,addr2>] [--policy <policy_file>]

Batch execution:

run-batch <program_dir> [--recursive] [--policy <policy_file>] [--output-dir <dir>]

Trace execution:

trace <program_file> [--trace-level basic|full|policy] [--output <trace_file>]

Inspect VM state:

inspect <program_file> [--state registers|memory|stack] [--address <addr>]

Examples

# Basic program execution
run hello_world.t81

# Run with policy enforcement and model weights
run inference.t81 --policy security.apl --weights-model model.t81w

# Debug with breakpoints
debug complex_program.t81 --breakpoints 0x1000,0x2000 --policy debug.apl

# Full execution trace
trace inference.t81 --trace-level full --output execution_trace.json

# Batch process all programs
run-batch ./programs/ --recursive --policy batch.apl --output-dir ./results/

# Inspect VM registers after execution
inspect program.t81 --state registers

Execution Modes

Standard Mode:

• Deterministic execution with policy enforcement
• Bit-identical results for same inputs
• Complete audit trail generation

Debug Mode:

• Step-through execution with breakpoints
• Register and memory inspection
• Policy violation analysis

Trace Mode:

• Detailed execution flow recording
• Policy decision logging
• Performance metrics collection

Output Format

Successful Execution:

{
  "status": "completed",
  "program": "inference.t81",
  "execution_id": "CanonHash81...",
  "timestamp": "2026-04-04T15:30:00Z",
  "result": {
    "output": "tensor_result_data",
    "exit_code": 0,
    "determinism_hash": "CanonHash81..."
  },
  "policy": {
    "enforced": true,
    "violations": 0,
    "decisions_made": 1247
  },
  "performance": {
    "instructions_executed": 5421,
    "execution_time_ms": 23,
    "memory_peak_bytes": 1048576
  }
}

Policy Violation:

{
  "status": "policy_violation",
  "program": "unauthorized.t81",
  "execution_id": "CanonHash81...",
  "timestamp": "2026-04-04T15:30:00Z",
  "error": {
    "type": "policy_denial",
    "instruction": 0x1234,
    "reason": "unauthorized_resource_access",
    "policy_rule": "deny_file_write"
  },
  "policy": {
    "enforced": true,
    "violations": 1,
    "decision_point": "instruction_0x1234"
  }
}

Supported File Types

T81Lang Source (.t81):

• High-level ternary language programs
• Requires compilation before execution

TISC Bytecode (.tisc):

• Compiled T81 instruction set
• Direct execution by T81VM

Bundle Files (.bundle):

• Decision bundles with provenance
• Executed with bundle consumption contract

Determinism Guarantees

• Bit-identical execution: Same inputs produce identical outputs
• Reproducible traces: Execution logs are deterministic
• Policy consistency: Same policy decisions for identical states
• Temporal determinism: Execution timing is predictable

Error Handling

• Compilation errors: Clear syntax and semantic error reporting
• Runtime errors: Detailed exception information with stack traces
• Policy violations: Specific rule violations with remediation suggestions
• Resource errors: Memory and resource limit handling

Security Notes

• All execution subject to Axion policy validation
• No side effects without policy approval
• Complete audit trail for compliance
• Sandboxed execution environment

Integration

This skill wraps the T81VM execution engine:

t81 vm run <program> --policy <policy> --trace --json