audit-telemetry

name	audit-telemetry
description	Analyze local telemetry logs to produce metrics on skill usage, agent invocations, and workflow patterns.
metadata	{"author":"cascadian-gamers","version":"1.0"}

Audit Telemetry

Analyze local telemetry logs to produce metrics on skill usage, agent invocations, and workflow patterns.

Input

Optional: date range (default: last 7 days), focus area (e.g., "reviews", "subagents", "decisions").

Process

Phase 1: Load Logs

1. List all .jsonl files in .kiro/telemetry/.
2. Filter to the requested date range.
3. Parse all JSON lines into events.

Phase 2: Compute Metrics

Skill Usage:

Metric	How
Total skill invocations	Count type: "skill" events
Skills by frequency	Group by skill, count, sort descending
Average duration per skill	Mean duration_sec for completed events, grouped by skill
Skill success rate	completed / (completed + failed) per skill
Most common trigger	user vs auto (chained) ratio

Subagent Usage:

Metric	How
Total subagent invocations	Count type: "subagent" events
Agents by frequency	Group by agent, count
Average findings per review	Mean findings for review agents
Subagent failure rate	failed / total per agent

Review Decisions:

Metric	How
Total decisions	Count type: "decision" events
Agree/Disagree/Defer ratio	Group by assessment, count
Most disagreed categories	Group disagreements by skill/category

Workflow Patterns:

Metric	How
Skill chains	Count type: "chain" events, show most common chains
Average review iterations	Count review→fix loops per task
Time from implement start to commit	Duration between first started and last completed for a task

Phase 3: Compliance Check

Flag violations:

• Skills that ran without a started event (missing start log)
• Skills that have started but no completed/failed (abandoned or forgot to log)
• Subagent invocations without a parent skill (orphaned)
• Review cycles with zero decision events (rubber-stamped)

Phase 4: Present Report

📊 Telemetry Report: 2026-02-19 to 2026-02-25

SKILL USAGE (last 7 days):
| Skill | Invocations | Avg Duration | Success Rate |
|-------|-------------|-------------|--------------|
| implement-and-review-loop | 8 | 7.2 min | 100% |
| review-code | 8 | 1.5 min | 87% |
| session-handoff | 5 | 0.5 min | 100% |
| validate-deployment | 3 | 2.1 min | 100% |
| push-and-pr | 3 | 0.8 min | 100% |

SUBAGENT USAGE:
| Agent | Invocations | Avg Findings | Failure Rate |
|-------|-------------|-------------|--------------|
| review-security | 8 | 2.1 | 0% |
| review-maintainability | 8 | 3.4 | 12% |
| review-test-quality | 8 | 1.8 | 0% |
| review-infrastructure | 8 | 0.9 | 0% |

REVIEW DECISIONS:
  Agree: 12 (35%) | Disagree: 18 (53%) | Defer: 4 (12%)
  Top disagreement: "DRY violation" (5 times)

COMPLIANCE:
  ✅ All skills logged start + end
  ⚠️ 1 subagent invocation without parent skill (2026-02-22)
  ✅ All review cycles have decision events

Rules

• Read-only — this skill never modifies telemetry files.
• Present metrics, don't judge. Let The Brougham 22 draw conclusions.
• Flag compliance violations clearly — these indicate the logging rules aren't being followed.
• If no telemetry files exist, say so and remind that .kiro/steering/telemetry.md requires logging.
• Refer to the user as "The Brougham 22".