t81-ai-tasks

name	t81_ai_tasks
description	Execute bounded AI task chains (assess-fixed, route-fixed, classify-fixed) with deterministic guarantees
metadata	{"openclaw":{"os":["darwin","linux"],"requires":{"bins":["t81"]}}}

T81 AI Tasks Skill

This skill executes bounded AI task chains that provide deterministic decision-making with complete provenance and policy enforcement.

Usage

When the user needs to run structured AI decision workflows with guaranteed determinism, use this skill to execute bounded task chains.

Commands

Assess-Fixed Task:

assess <input_data> [--model <model_file>] [--policy <policy_file>] [--output <output_file>]

Route-Fixed Task:

route <input_data> [--routes <route_config>] [--policy <policy_file>] [--output <output_file>]

Classify-Fixed Task:

classify <input_data> [--classes <class_config>] [--policy <policy_file>] [--output <output_file>]

Chain Execution:

chain <chain_config> [--input <input_data>] [--policy <policy_file>] [--bundle-output <bundle_file>]

Task Status:

status <task_id> [--detailed] [--provenance]

Task History:

history [--limit <count>] [--filter assess|route|classify] [--json]

Examples

# Assess fixed decision with model
assess market_data.csv --model trading_model.t81w --policy risk_assessment.apl

# Route input through predefined decision paths
route customer_request.json --routes routing_config.json --policy compliance.apl

# Classify data with fixed categories
classify medical_data.json --classes diagnosis_classes.json --policy medical.apl

# Execute complete decision chain
chain trading_workflow.json --input market_data.csv --bundle-output decisions.bundle

# Check task execution status
status task_12345 --detailed --provenance

# View task execution history
history --limit 10 --filter assess --json

Task Types

Assess-Fixed:

• Evaluates input against fixed criteria
• Produces deterministic assessment results
• Used for risk analysis, compliance checking
• Complete audit trail for decisions

Route-Fixed:

• Routes inputs through predetermined paths
• Deterministic routing decisions
• Used for workflow orchestration
• Policy-gated path selection

Classify-Fixed:

• Classifies inputs into fixed categories
• Deterministic classification results
• Used for categorization and labeling
• Consistent classification across runs

Chain Composition

Sequential Chains:

Input → Assess → Route → Classify → Output

Parallel Chains:

Input → [Assess1, Assess2, Assess3] → Aggregate → Output

Conditional Chains:

Input → Assess → Route(Condition) → [ClassifyA|ClassifyB] → Output

Configuration Format

Task Configuration:

{
  "task_type": "assess-fixed",
  "model": "model.t81w",
  "policy": "security.apl",
  "parameters": {
    "confidence_threshold": 0.95,
    "determinism_mode": "strict"
  },
  "output_format": "bundle"
}

Chain Configuration:

{
  "chain_id": "trading_workflow",
  "tasks": [
    {
      "type": "assess-fixed",
      "model": "risk_model.t81w",
      "policy": "risk_assessment.apl"
    },
    {
      "type": "route-fixed", 
      "routes": "trading_routes.json",
      "policy": "execution_policy.apl"
    },
    {
      "type": "classify-fixed",
      "classes": "trade_classes.json",
      "policy": "classification.apl"
    }
  ],
  "bundle_output": true
}

Output Format

Task Result:

{
  "task_id": "CanonHash81...",
  "task_type": "assess-fixed",
  "status": "completed",
  "timestamp": "2026-04-04T15:30:00Z",
  "input_hash": "CanonHash81...",
  "result": {
    "assessment": "high_risk",
    "confidence": 0.98,
    "determinism_verified": true
  },
  "provenance": {
    "model_hash": "CanonHash81...",
    "policy_hash": "CanonHash81...",
    "execution_trace": "CanonHash81..."
  },
  "bundle": {
    "created": true,
    "bundle_hash": "CanonHash81...",
    "bundle_location": "/bundles/assessment_123.bundle"
  }
}

Chain Result:

{
  "chain_id": "trading_workflow",
  "status": "completed",
  "timestamp": "2026-04-04T15:30:00Z",
  "tasks_executed": 3,
  "tasks_completed": 3,
  "tasks_failed": 0,
  "final_result": {
    "decision": "execute_trade",
    "parameters": {
      "symbol": "AAPL",
      "quantity": 100,
      "confidence": 0.97
    }
  },
  "bundle": {
    "bundle_hash": "CanonHash81...",
    "contains_provenance": true,
    "determinism_guaranteed": true
  }
}

Determinism Guarantees

• Input Determinism: Same inputs produce identical outputs
• Temporal Determinism: Execution timing is reproducible
• Policy Determinism: Same policy decisions for identical states
• Chain Determinism: Complete chain execution is reproducible

Policy Integration

Pre-Execution Checks:

• Input validation against policies
• Model authorization verification
• Resource access permissions

Runtime Enforcement:

• Step-by-step policy validation
• Resource usage monitoring
• Decision point auditing

Post-Execution Validation:

• Output compliance checking
• Bundle integrity verification
• Provenance completeness

Error Handling

• Task Failures: Detailed error information with recovery suggestions
• Policy Violations: Specific rule violations with remediation steps
• Model Errors: Loading and execution problem handling
• Chain Breaks: Partial completion handling and rollback

Security Notes

• All tasks subject to Axion policy enforcement
• Complete audit trail for regulatory compliance
• Immutable bundle storage for provenance
• Deterministic execution prevents manipulation

Integration

This skill wraps T81's bounded AI task chains:

t81 ai task <task_type> --input <data> --model <model> --policy <policy> --bundle