Machinery Diagnostic Agent Guardrail Brief

Role: Principal Reliability Engineering Director leading maintenance safety standards and field diagnostic systems.

Context

• Equipment family: {{machinery_fleet_type}}
• Live and historical data inputs: {{telemetry_data_streams}}
• Agent operational envelope: {{diagnostic_model_scope}}
• Prohibited outputs and commands: {{unauthorized_command_patterns}}
• Target user profile: {{technician_skill_tier}}
• Emergency support channel: {{incident_escalation_path}}

Task

Author a diagnostic safety guardrail brief detailing response filtering, telemetry ground truth verification, and technician safety rails for AI agents troubleshooting {{machinery_fleet_type}}.

Method

1. Scope the diagnostic remit of the agent against the specific components of {{machinery_fleet_type}}.
2. Define ground truth verification steps matching agent assertions against live metrics from {{telemetry_data_streams}}.
3. Formulate strict hallucination barriers that prevent the generation of unverified maintenance steps outside {{diagnostic_model_scope}}.
4. Construct negative pattern matching filters to intercept and neutralize {{unauthorized_command_patterns}}.
5. Tailor output clarity, danger warnings, and procedural complexity to {{technician_skill_tier}}.
6. Specify fallback behavior when sensor confidence drops or diagnostic ambiguity exceeds safety tolerances.
7. Map immediate handoff mechanisms directly to {{incident_escalation_path}} for high-voltage, pressure, or kinetic hazards.
8. Establish telemetry feedback loops to capture near-miss advisory anomalies.

Constraints

• The diagnostic agent MUST NOT recommend maintenance steps that bypass lockout/tagout (LOTO) procedures.
• The agent MUST explicitly display confidence ratings and underlying telemetry references for every repair step.
• Output instructions must match the technical vocabulary and certification tier of {{technician_skill_tier}}.
• Every prohibited command in {{unauthorized_command_patterns}} must have an automated blocking rule.

Output format

A technical guardrail brief comprising:

• System Envelope and Advisory Scope (150 words)
• Telemetry Grounding and Assertion Rules (table linking {{telemetry_data_streams}} to verification checks)
• Hallucination and Unsafe Output Filters (pattern list and blocking actions for {{unauthorized_command_patterns}})
• Technician Safety and LOTO Enforcements (mandatory safety warnings and confirmation steps)
• Rapid Escalation and Human Handoff Procedures (flowchart-style steps pointing to {{incident_escalation_path}})

Self-review

• Ensure no scenario permits the agent to suggest bypassing physical machine interlocks.
• Confirm that grounding rules verify incoming data against {{telemetry_data_streams}} before providing advice.
• Validate that all prohibited command patterns in {{unauthorized_command_patterns}} are covered with zero ambiguity.