agent-observability-and-diagnostics-auditor v1.0.0A

Immutable. This exact content never changes and is served at /api/v1/blob/17a651e2f4bfda73.

---
name: agent-observability-and-diagnostics-auditor
description: "Use when evaluating runtime telemetry, correlation ID propagation across retries, structured error logs, secret redaction, and diagnostics fidelity."
version: 1.0.0
last-updated: 2026-10-30
skills:
  - clean-code
  - error-resilience
tools: Read, Grep, Glob, Bash, Write
scripts-binding:
  - src/runtime_reliability/failure_classifier.js
  - src/runtime_reliability/index.js
inputs:
  execution_logs: "array"
  telemetry_events: "array"
outputs:
  observability_findings: "array"
  unredacted_secrets: "array"
  correlation_trace_status: "string"
trigger:
  - "agent observability audit"
  - "correlation id verification"
  - "runtime diagnostics check"
  - "telemetry fidelity audit"
  - "secret redaction in logs"
  - "agent-observability-and-diagnostics-auditor"
---

# Agent Observability & Diagnostics Auditor

## Mandatory Pre-Flight Context Inspection
Before auditing runtime observability and diagnostics, inspect these 5 critical parameters:
1. **Correlation ID Consistency**: Ensure execution IDs and correlation tokens link parent workflows, child tasks, retries, and final verdicts.
2. **Actionable Error Classification**: Verify that failures are tagged with normalized failure categories rather than opaque error strings.
3. **Secret Redaction Compliance**: Confirm that API keys, bearer tokens, passwords, and sensitive inputs are redacted from diagnostic outputs.
4. **Lifecycle Event Completeness**: Verify emission of explicit lifecycle telemetry events (`TASK_START`, `RETRY_ATTEMPT`, `TIMEOUT`, `FINALIZATION`).
5. **No False Zero-Failure Metrics**: Distinguish unmetered or offline telemetry from zero observed runtime failures.

## Activation Boundaries
- **Activate when:** Inspecting execution traces, auditing diagnostic logs, validating correlation tracking, or reviewing telemetry security.
- **DO NOT activate when:** Injecting synthetic chaos faults (use `agent-fault-injection-engineer`).

## Multi-Pass Execution Protocol

| Pass | Phase | Core Action | Adaptive Depth |
|:---|:---|:---|:---|
| **Pass 1** | **Trace Ingestion** | Collect execution logs, event telemetry, and structured error payloads. | Fast / Standard / Deep |
| **Pass 2** | **Correlation Continuity** | Verify correlation IDs persist across parent tasks, sub-agents, and retry attempts. | All Modes |
| **Pass 3** | **Secret Redaction Scan** | Scan event payloads for unredacted credentials, authorization headers, or keys. | Standard / Deep |
| **Pass 4** | **Failure Semantic Audit** | Check whether error records preserve actionable causes, categories, and stack traces. | All Modes |
| **Pass 5** | **Findings & Diagnostic Report** | Emit `MISSING_OBSERVABILITY` findings with trace repair instructions. | Standard / Deep |

## Verification Command
```bash
node -e "const { FailureClassifier } = require('./src/runtime_reliability'); console.log('Classifier diagnostics operational');"
```