logvalet-intelligence

lv activity stats と lv project health を材料に、LLM がアクティビティパターンの偏り・異常を解釈してリスク評価を生成する。

For full logvalet CLI documentation, see the logvalet skill.

When to use this skill

Use logvalet-intelligence when you need to:

detect unusual spikes or drops in project activity
identify team members with disproportionate contribution (over- or under-active)
surface hidden risks that raw stats do not make explicit
evaluate team health from an activity perspective
support retrospectives with data-driven insights

Workflow

Step 1: Identify project

If the user provides a project key, use it directly.

If not provided, list available projects:

lv project list -f md

Then ask the user to select one.

Step 2: Determine scope and period

Ask in a single question (if not already specified):

--scope: 集計スコープ (project / user / space, デフォルト: project)
--since / --until: 集計期間（ISO 8601 形式）
--top-n: 上位表示数（デフォルト: 5）

If the user wants a quick overview, use defaults.

Step 3: Fetch materials in parallel

Run both commands in parallel:

lv activity stats --scope project -k PROJECT_KEY --since YYYY-MM-DDT00:00:00Z --until YYYY-MM-DDT23:59:59Z --top-n 10 -f json

lv project health PROJECT_KEY -f json

The activity stats output includes:

total_count: 期間内総アクティビティ数
by_type: アクティビティタイプ別内訳
by_actor: アクター（ユーザー）別内訳
by_hour: 時間帯別分布
by_day_of_week: 曜日別分布
top_active_actors: 最も活発なアクター上位 N 件
top_active_types: 最も多いタイプ上位 N 件

Step 4: Analyze patterns and detect anomalies

Using the stats and health data, reason about:

Activity volume:

Is total activity count unusually high or low for the period?
Are there unexpected spikes or gaps?

Actor distribution:

Is activity heavily concentrated on a few members? (Gini-like inequality)
Are any team members completely absent from recent activity?
Is there an unexpectedly high single-actor share (>60%)?

Type distribution:

Is there an unusual imbalance between issue creation vs. resolution?
Are comment activities disproportionately high (discussion-heavy, no resolution)?
Are there unexpected activity types?

Cross-signal correlation:

Compare activity stats with project health blockers/stale issues
If stale issues are high but activity is also high → activity may not be on the right issues
If activity is low but no blockers → team may be in a quiet phase (confirm intentionality)

Step 5: Present intelligence report

## アクティビティインテリジェンスレポート — PROJECT_KEY

> 分析期間: FROM〜TO / 総アクティビティ: N件 / 生成日時: YYYY-MM-DD

---

### アクティビティ概要

| 指標 | 値 |
|------|-----|
| 総アクティビティ数 | N |
| アクター数 | N |
| 最も多いタイプ | TypeName (N%) |
| 最も活発なメンバー | UserName (N件, N%) |

---

### 異常・偏り検出

#### 偏り指標
- **アクター集中度:** <低/中/高> — 上位1名が全体の N% を占める
- **タイプバランス:** <問題なし / 課題作成偏重 / コメント偏重 / ...>

#### 検出された異常
1. **<異常の名称>**
   - 観測値: <具体的な数値>
   - 解釈: <何が起きている可能性があるか>
   - リスク: <低/中/高>

2. ...（なければ「異常なし」）

---

### ヘルス相関分析

<activity stats と project health の相関から導かれる洞察>

---

### 推奨アクション

1. <最優先のアクション>
2. <次に重要なアクション>
3. ...（最大5件）

---

**サマリー:**
- 全体リスクレベル: <低/中/高>
- 主な懸念事項: <1-2行>

Step 6: No writes

This is a read-only skill. No issue updates are performed.

If the user wants to act on findings, switch to the logvalet skill for updates, or logvalet-triage for issue triage.

Notes

activity stats と project health は並列実行で取得する（API 呼び出し最小化）
アクター集中度の高低目安: 上位1名が全体の >60% → 高、30-60% → 中、<30% → 低
--top-n 10 を推奨（デフォルト5では偏りが見えにくい場合がある）
期間を指定しない場合、Backlog API の直近アクティビティ（最大100件）が返る

Anti-patterns

Do not auto-update issues based on intelligence findings — always present findings first
Do not interpret low activity as a problem without context (planned quiet phase, holidays, etc.)
Do not skip the project identification step — never guess the project key
Do not run both commands sequentially when they can run in parallel

logvalet:intelligence