Skip to content
Back to skills

Telemetry Analysis

ASecurity

Use when asked to analyze hook activity, token usage, audit logs, or agent behavior patterns. Triggers on: 'xem log', 'hook nào fire nhiều', 'token usage', 'audit trail', 'what hooks fired', 'session summary', 'telemetry report', 'hook health', 'agent activity'.

  • 10 stars
  • 0 votes
  • 0 copies
  • 5 views
  • Added September 9, 2026
ai-agentsrustbash

Security analysis

A100/100

Scanned September 9, 2026

npx -y skills add yanacuti1121/Yana-AI --skill telemetry-analysis --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Telemetry Analysis?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Telemetry Analysis
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/yanacuti1121-telemetry-analysis/badge)](https://www.skillsdirectory.com/skills/yanacuti1121-telemetry-analysis)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: telemetry-analysis
description: "Use when asked to analyze hook activity, token usage, audit logs, or agent behavior patterns. Triggers on: 'xem log', 'hook nào fire nhiều', 'token usage', 'audit trail', 'what hooks fired', 'session summary', 'telemetry report', 'hook health', 'agent activity'."
---

# Telemetry Analysis Skill

Đọc và phân tích dữ liệu telemetry local của YAMTAM để hiểu sức khỏe hệ thống,
pattern hook, và chi phí token — không cần network, không cần external service.

## Nguồn dữ liệu

| File | Nội dung | Format |
|------|---------|--------|
| `.claude/state/audit-chain.log` | Mọi tool call theo thứ tự thời gian | JSONL, hash-chain |
| `.claude/state/telemetry.jsonl` | Chi tiết từng hook: tool, decision, duration | JSONL |
| `.claude/state/session-trust.json` | Trust score hiện tại + lịch sử decrement | JSON |
| `.claude/session-read-log.txt` | Files đã đọc trong session | Plain text |

## Workflow phân tích

### Step 1 — Kiểm tra file tồn tại

```bash
ls -la .claude/state/ 2>/dev/null || echo "No state dir — YAMTAM not applied yet"
wc -l .claude/state/audit-chain.log 2>/dev/null || echo "No audit log"
wc -l .claude/state/telemetry.jsonl 2>/dev/null || echo "No telemetry"
```

### Step 2 — Hook activity summary

```bash
# Hook nào fire nhiều nhất
jq -r '.hook // "unknown"' .claude/state/telemetry.jsonl 2>/dev/null \
  | sort | uniq -c | sort -rn | head -10

# Tỉ lệ block vs allow vs warn
jq -r '.decision // "unknown"' .claude/state/telemetry.jsonl 2>/dev/null \
  | sort | uniq -c | sort -rn
```

### Step 3 — Tool call pattern

```bash
# Tool nào được gọi nhiều nhất
jq -r '.tool // "unknown"' .claude/state/audit-chain.log 2>/dev/null \
  | sort | uniq -c | sort -rn | head -10

# Timeline: 10 tool call gần nhất
jq -r '[.ts, .tool, .agent] | @tsv' .claude/state/audit-chain.log 2>/dev/null \
  | tail -10
```

### Step 4 — Trust score health

```bash
cat .claude/state/session-trust.json 2>/dev/null | jq '.'
```

Diễn giải:
- Score ≥ 80: Healthy — agent hoạt động tốt
- Score 50–79: Caution — một số unverified claim
- Score < 50: 🔴 LOW TRUST — require double evidence cho mọi claim

### Step 5 — Expensive operations

```bash
# Hook nào tốn thời gian nhất (nếu có duration field)
jq -r 'select(.duration_ms != null) | [.duration_ms, .hook] | @tsv' \
  .claude/state/telemetry.jsonl 2>/dev/null \
  | sort -rn | head -10
```

### Step 6 — Bypass usage

```bash
# Có env var bypass nào được dùng không
jq -r 'select(.input | strings | contains("YAMTAM_BYPASS") or contains("APPROVED")) | .input' .claude/state/audit-chain.log 2>/dev/null | sort | uniq -c
```

## Report format

Sau khi chạy tất cả steps, xuất report:

```
=== Telemetry Report — [date] ===

Hook activity:
  Most fired: [hook] (N times)
  Block rate:  N% | Warn rate: N% | Allow rate: N%

Tool calls:
  Most used: [tool] (N times)
  Total calls: N

Trust score: N/100 — [HEALTHY | CAUTION | LOW TRUST]
  Decrements this session: N

Top expensive hooks: [list if available]

Bypass usage: [N bypasses detected | None]

Recommendations:
  - [nếu block rate cao: xem xét hook đang quá strict]
  - [nếu trust thấp: review claim verbs trong session]
  - [nếu hook im lặng: kiểm tra hook có được wire đúng không]
```

## Khi không có dữ liệu

Nếu `.claude/state/` không tồn tại:
1. YAMTAM chưa được apply vào project này
2. Hoặc agent đang chạy từ engine repo, không phải target project
3. Hướng dẫn: `unzip releases/yana-ai-latest.zip -d .claude/`

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.