insights

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Translation

Chinese

Insights Skill

When to Use This Skill

何时使用此Skill

IMPORTANT: You MUST activate this skill when you detect ANY of the following signals from the user. Do NOT ask for confirmation — just start executing the orchestration steps below.

重要提示：您必须激活此Skill，当您检测到用户发出以下任一信号时。无需请求确认——直接开始执行下方的编排步骤。

Trigger Phrases (any language)

触发短语（任意语言）

Use this skill when the user:

Asks to analyze sessions: "analyze my sessions", "세션 분석해줘", "look at my sessions"
Asks for a report: "generate a report", "리포트 만들어줘", "give me a report"
Asks about usage patterns: "how have I been using you?", "사용 패턴 보여줘", "what are my patterns?"
Asks about statistics: "show my stats", "통계 보여줘", "how many sessions?"
Mentions workflow insights: "insights on my workflow", "워크플로우 분석", "what's working well?"
Asks about productivity: "am I productive?", "내 생산성은?", "how efficient am I?"
Uses keywords: "insights", "인사이트", "analytics", "분석", "report", "리포트", "stats", "통계", "sessions", "세션"
Asks vaguely about history: "what have we been doing?", "그동안 뭐했지?", "show me my history"

当用户有以下需求时使用此Skill：

要求分析会话："analyze my sessions"、"세션 분석해줘"、"look at my sessions"
要求生成报告："generate a report"、"리포트 만들어줘"、"give me a report"
询问使用模式："how have I been using you?"、"사용 패턴 보여줘"、"what are my patterns?"
询问统计数据："show my stats"、"통계 보여줘"、"how many sessions?"
提及工作流洞察："insights on my workflow"、"워크플로우 분석"、"what's working well?"
询问生产力："am I productive?"、"내 생산성은?"、"how efficient am I?"
使用关键词："insights"、"인사이트"、"analytics"、"분석"、"report"、"리포트"、"stats"、"통계"、"sessions"、"세션"
模糊询问历史记录："what have we been doing?"、"그동안 뭐했지?"、"show me my history"

What NOT to Trigger On

请勿触发的场景

Do NOT use this skill when the user:

Asks about a specific coding task (not session analysis)
Wants to read a single session's content (use session tools directly)
Asks about this skill's code itself (just read the files)

当用户有以下需求时，请勿使用此Skill：

询问特定编码任务（非会话分析）
想要查看单个会话的内容（直接使用会话工具）
询问此Skill的代码本身（直接读取文件即可）

Overview

概述

The insights skill generates a comprehensive HTML report analyzing your AI coding assistant usage patterns. It automatically detects which CLI you're using (Claude Code, OpenCode, or Codex), collects session data, performs 8 categories of analysis, and produces a styled HTML report with actionable insights.

What it produces:

HTML report with 8 analysis sections
Quantitative statistics (sessions, messages, duration, tools used)
Qualitative insights (workflow patterns, friction points, suggestions)
CLI-specific recommendations
Self-contained file:// URL (no external dependencies)

Insights Skill会生成一份全面的HTML报告，分析您的AI编码助手使用模式。它会自动检测您使用的CLI（Claude Code、OpenCode或Codex），收集会话数据，执行8类分析，并生成带有可操作洞察的样式化HTML报告。

生成内容包括：

包含8个分析板块的HTML报告
量化统计数据（会话数、消息数、时长、使用的工具）
定性洞察（工作流模式、痛点、建议）
特定CLI的建议
独立的file:// URL（无外部依赖）

Prerequisites

前置条件

Required:

bash 4+ (check:
```
bash --version
```
)
jq (JSON processor)

Supported CLIs:

Claude Code 2.1+
OpenCode 1.1+
Codex

Verify dependencies:

bash

./scripts/check-deps.sh

If dependencies are missing, the script provides install instructions.

必需项：

bash 4+（检查：
```
bash --version
```
）
jq（JSON处理器）

支持的CLI：

Claude Code 2.1+
OpenCode 1.1+
Codex

验证依赖项：

bash

./scripts/check-deps.sh

如果缺少依赖项，脚本会提供安装说明。

Orchestration Steps

编排步骤

Follow these steps in order. Each step depends on the previous one.

请按顺序执行以下步骤，每一步都依赖于上一步的结果。

Step 1: Check Dependencies

步骤1：检查依赖项

Run:

scripts/check-deps.sh

Purpose: Verify jq and bash 4+ are installed.

Action:

If exit code 0: Continue to Step 2
If exit code 1: Show user the install instructions from stderr, then abort

Example:

bash

cd skills/insights
./scripts/check-deps.sh

执行命令：

scripts/check-deps.sh

目的： 验证jq和bash 4+是否已安装。

操作：

如果退出码为0：继续执行步骤2
如果退出码为1：向用户显示stderr中的安装说明，然后终止流程

示例：

bash

cd skills/insights
./scripts/check-deps.sh

Step 2: Detect CLI Type

步骤2：检测CLI类型

Run:

scripts/detect-cli.sh

Purpose: Auto-detect which AI coding CLI the user is running.

Output: Single line:

claude-code

opencode

codex

, or

unknown

Action:

Capture the output as
```
$CLI_TYPE
```
If output is
```
unknown
```
: Ask user which CLI they're using, or abort
Otherwise: Continue to Step 3

Example:

bash

CLI_TYPE=$(./scripts/detect-cli.sh)
echo "Detected CLI: $CLI_TYPE"

Override: User can specify CLI manually:

bash

CLI_TYPE=$(./scripts/detect-cli.sh --cli claude-code)

执行命令：

scripts/detect-cli.sh

目的： 自动检测用户正在运行的AI编码CLI。

输出： 单行内容：

claude-code

、

opencode

、

codex

或

unknown

操作：

将输出捕获为
```
$CLI_TYPE
```
如果输出为
```
unknown
```
：询问用户使用的CLI类型，或终止流程
否则：继续执行步骤3

示例：

bash

CLI_TYPE=$(./scripts/detect-cli.sh)
echo "Detected CLI: $CLI_TYPE"

手动覆盖： 用户可以手动指定CLI：

bash

CLI_TYPE=$(./scripts/detect-cli.sh --cli claude-code)

Step 3: Collect Session Metadata

步骤3：收集会话元数据

Run:

scripts/collect-sessions.sh --cli $CLI_TYPE

Purpose: Collect session metadata from the CLI's storage directory.

Output: JSON array of session objects (stdout)

Action:

Capture output as
```
$SESSIONS_JSON
```

Verify it's valid JSON:

echo "$SESSIONS_JSON" | jq . > /dev/null

If empty array
```
[]
```
: Inform user no sessions found, abort
Otherwise: Continue to Step 4

Example:

bash

SESSIONS_JSON=$(./scripts/collect-sessions.sh --cli "$CLI_TYPE" --limit 50)
echo "$SESSIONS_JSON" | jq 'length'  # Show session count

Options:

```
--limit N
```
: Limit to N most recent sessions (default: 50)
```
--session-dir <path>
```
: Override default session directory

执行命令：

scripts/collect-sessions.sh --cli $CLI_TYPE

目的： 从CLI的存储目录收集会话元数据。

输出： 会话对象的JSON数组（标准输出）

操作：

将输出捕获为
```
$SESSIONS_JSON
```

验证其为有效JSON：

echo "$SESSIONS_JSON" | jq . > /dev/null

如果为空数组
```
[]
```
：告知用户未找到会话，终止流程
否则：继续执行步骤4

示例：

bash

SESSIONS_JSON=$(./scripts/collect-sessions.sh --cli "$CLI_TYPE" --limit 50)
echo "$SESSIONS_JSON" | jq 'length'  # 显示会话数量

可选参数：

```
--limit N
```
：限制为最近的N个会话（默认值：50）
```
--session-dir <path>
```
：覆盖默认会话目录

Step 4: Aggregate Statistics

步骤4：聚合统计数据

Run:

scripts/aggregate-stats.sh

(reads from stdin)

Purpose: Compute quantitative statistics from session metadata.

Input: Session metadata JSON array (from Step 3)

Output: Aggregated stats JSON object (stdout)

Action:

Pipe
```
$SESSIONS_JSON
```
to aggregate-stats.sh
Capture output as
```
$STATS_JSON
```
Verify valid JSON
Continue to Step 5

Example:

bash

STATS_JSON=$(echo "$SESSIONS_JSON" | ./scripts/aggregate-stats.sh)
echo "$STATS_JSON" | jq '.total_sessions, .total_messages'

Stats included:

total_sessions, total_messages, total_duration_hours
date_range (start, end)
tool_counts, languages, projects
git_commits, git_pushes
days_active, messages_per_day
message_hours (24-element array)

执行命令：

scripts/aggregate-stats.sh

（从标准输入读取数据）

目的： 根据会话元数据计算量化统计数据。

输入： 会话元数据JSON数组（来自步骤3）

输出： 聚合后的统计数据JSON对象（标准输出）

操作：

将
```
$SESSIONS_JSON
```
通过管道传递给aggregate-stats.sh
将输出捕获为
```
$STATS_JSON
```
验证为有效JSON
继续执行步骤5

示例：

bash

STATS_JSON=$(echo "$SESSIONS_JSON" | ./scripts/aggregate-stats.sh)
echo "$STATS_JSON" | jq '.total_sessions, .total_messages'

包含的统计数据：

total_sessions、total_messages、total_duration_hours
date_range（开始时间、结束时间）
tool_counts、languages、projects
git_commits、git_pushes
days_active、messages_per_day
message_hours（24元素数组）

Step 5: Extract Session Facets (Optional but Recommended)

步骤5：提取会话维度（可选但推荐）

Purpose: Extract qualitative facets from individual sessions for richer analysis.

Limit: Up to 20 sessions (to manage token cost)

Prompt: Use the facet extraction prompt from

references/facet-extraction.md

Action:

Select up to 20 sessions from
```
$SESSIONS_JSON
```
(prioritize recent, diverse sessions)
For each session:
- Read the full session data (messages, tool calls, outcomes)
- Execute the facet extraction prompt with session context
- Parse the JSON response
Aggregate facets into
```
$FACETS_JSON
```

Facet fields:

goal_categories (debugging, feature_implementation, refactoring, etc.)
user_satisfaction_counts (happy, satisfied, frustrated, etc.)
friction_counts (misunderstood_request, buggy_code, etc.)
outcome_level (success, partial_success, blocked, abandoned)

Cache: Store extracted facets in

~/.agent-insights/cache/facets/<session-id>.json

to avoid re-extraction.

Example:

bash

undefined

目的： 从单个会话中提取定性维度，以获得更丰富的分析结果。

限制： 最多20个会话（以控制令牌成本）

提示词： 使用

references/facet-extraction.md

中的维度提取提示词

操作：

从
```
$SESSIONS_JSON
```
中选择最多20个会话（优先选择近期、多样化的会话）
对于每个会话：
- 读取完整会话数据（消息、工具调用、结果）
- 结合会话上下文执行维度提取提示词
- 解析JSON响应
将维度聚合为
```
$FACETS_JSON
```

维度字段：

goal_categories（调试、功能实现、重构等）
user_satisfaction_counts（满意、较满意、沮丧等）
friction_counts（请求被误解、代码有bug等）
outcome_level（成功、部分成功、受阻、已放弃）

缓存： 将提取的维度存储在

~/.agent-insights/cache/facets/<session-id>.json

中，避免重复提取。

示例：

bash

undefined

Pseudo-code (agent executes this logic)

伪代码（由Agent执行此逻辑）

for session in (first 20 sessions): cache_file="~/.agent-insights/cache/facets/${session.session_id}.json" if cache_file exists: facet = read cache_file else: facet = execute_facet_extraction_prompt(session) write facet to cache_file facets.append(facet)

undefined

for session in (前20个会话): cache_file="~/.agent-insights/cache/facets/${session.session_id}.json" if cache_file exists: facet = 读取缓存文件 else: facet = 执行维度提取提示词(session) 将facet写入缓存文件 facets.append(facet)

undefined

Step 6: Run 8 Analysis Prompts

步骤6：运行8个分析提示词

Purpose: Generate qualitative insights across 8 categories.

Prompts: All 8 prompts are in

references/analysis-prompts.md

Context: Provide each prompt with:

```
$STATS_JSON
```
(aggregated statistics)
```
$FACETS_JSON
```
(extracted facets, if available)
```
$CLI_TYPE
```
(for CLI-specific suggestions)

Prompts to execute:

project_areas - What the user works on (4-5 areas)
interaction_style - How the user interacts (narrative + key_pattern)
what_works - Impressive workflows (3 items)
friction_analysis - Pain points (3 categories × 2 examples)
suggestions - CLI-agnostic + CLI-specific features (use
```
references/suggestions-by-cli.md
```
)
on_the_horizon - Future opportunities (3 items with copyable_prompt)
fun_ending - Memorable moment (headline + detail)
at_a_glance - Executive summary (4 fields: whats_working, whats_hindering, quick_wins, ambitious_workflows)

Execution:

Run all 8 prompts in parallel if possible (faster)
Each prompt returns a JSON object
Combine all 8 responses into
```
$INSIGHTS_JSON
```

Example:

json

{
  "project_areas": {...},
  "interaction_style": {...},
  "what_works": {...},
  "friction_analysis": {...},
  "suggestions": {...},
  "on_the_horizon": {...},
  "fun_ending": {...},
  "at_a_glance": {...}
}

Important: Each prompt includes "RESPOND WITH ONLY A VALID JSON OBJECT" instruction. Parse responses carefully.

目的： 生成8个类别的定性洞察。

提示词： 所有8个提示词都在

references/analysis-prompts.md

中

上下文： 为每个提示词提供以下内容：

```
$STATS_JSON
```
（聚合统计数据）
```
$FACETS_JSON
```
（提取的维度，如果有的话）
```
$CLI_TYPE
```
（用于特定CLI的建议）

需要执行的提示词：

project_areas - 用户的工作内容（4-5个领域）
interaction_style - 用户的交互方式（叙述+关键模式）
what_works - 有效的工作流（3项）
friction_analysis - 痛点（3个类别×2个示例）
suggestions - 独立于CLI和特定于CLI的功能（使用
```
references/suggestions-by-cli.md
```
）
on_the_horizon - 未来机会（3项，包含可复制的提示词）
fun_ending - 难忘时刻（标题+细节）
at_a_glance - 执行摘要（4个字段：whats_working、whats_hindering、quick_wins、ambitious_workflows）

执行方式：

如果可能，并行运行所有8个提示词（速度更快）
每个提示词返回一个JSON对象
将所有8个响应合并为
```
$INSIGHTS_JSON
```

示例：

json

{
  "project_areas": {...},
  "interaction_style": {...},
  "what_works": {...},
  "friction_analysis": {...},
  "suggestions": {...},
  "on_the_horizon": {...},
  "fun_ending": {...},
  "at_a_glance": {...}
}

重要提示： 每个提示词都包含“仅返回有效的JSON对象”的指令，请仔细解析响应。

Step 7: Generate HTML Report

步骤7：生成HTML报告

Run:

scripts/generate-report.sh --stats <stats-file> --cli $CLI_TYPE --output <path>

Purpose: Inject JSON data into HTML template and create final report.

Input:

```
$INSIGHTS_JSON
```
(from Step 6, via stdin)
```
$STATS_JSON
```
(from Step 4, via --stats file)
```
$CLI_TYPE
```
(via --cli flag)

Output: Path to generated HTML file (stdout)

Action:

Write
```
$STATS_JSON
```
to a temp file
Pipe
```
$INSIGHTS_JSON
```
to generate-report.sh
Capture output path as
```
$REPORT_PATH
```
Continue to Step 8

Example:

bash

STATS_FILE=$(mktemp)
echo "$STATS_JSON" > "$STATS_FILE"

REPORT_PATH=$(echo "$INSIGHTS_JSON" | ./scripts/generate-report.sh \
  --stats "$STATS_FILE" \
  --cli "$CLI_TYPE" \
  --language "${LANGUAGE:-en}" \
  --output ~/.agent-insights/reports/report-$(date +%Y-%m-%d).html)

echo "Report generated: $REPORT_PATH"

Default output:

./insights-report.html

Recommended output:

~/.agent-insights/reports/report-YYYY-MM-DD.html

执行命令：

scripts/generate-report.sh --stats <stats-file> --cli $CLI_TYPE --output <path>

目的： 将JSON数据注入HTML模板并生成最终报告。

输入：

```
$INSIGHTS_JSON
```
（来自步骤6，通过标准输入）
```
$STATS_JSON
```
（来自步骤4，通过--stats文件）
```
$CLI_TYPE
```
（通过--cli参数）

输出： 生成的HTML文件路径（标准输出）

操作：

将
```
$STATS_JSON
```
写入临时文件
将
```
$INSIGHTS_JSON
```
通过管道传递给generate-report.sh
将输出路径捕获为
```
$REPORT_PATH
```
继续执行步骤8

示例：

bash

STATS_FILE=$(mktemp)
echo "$STATS_JSON" > "$STATS_FILE"

REPORT_PATH=$(echo "$INSIGHTS_JSON" | ./scripts/generate-report.sh \
  --stats "$STATS_FILE" \
  --cli "$CLI_TYPE" \
  --language "${LANGUAGE:-en}" \
  --output ~/.agent-insights/reports/report-$(date +%Y-%m-%d).html)

echo "Report generated: $REPORT_PATH"

默认输出路径：

./insights-report.html

推荐输出路径：

~/.agent-insights/reports/report-YYYY-MM-DD.html

Step 8: Present to User

步骤8：向用户展示结果

Purpose: Show summary and provide file:// URL.

Action:

Read key stats from
```
$STATS_JSON
```
:
- total_sessions
- total_messages
- total_duration_hours
- date_range
Read at-a-glance summary from
```
$INSIGHTS_JSON
```
Present to user:

Example output:

✅ Insights Report Generated

📊 Summary:
- Sessions analyzed: 42
- Total messages: 1,247
- Time period: 2026-01-15 to 2026-02-06
- Total duration: 18.5 hours

🎯 At a Glance:
- What's working: You're effectively using Claude for rapid prototyping...
- What's hindering: Frequent context resets when switching between projects...
- Quick wins: Try using /compact to reduce token usage...
- Ambitious workflows: Explore multi-file refactoring with LSP tools...

📄 Full report: file://$REPORT_PATH

Open the report in your browser to see detailed insights across 8 categories.

目的： 显示摘要并提供file:// URL。

操作：

从
```
$STATS_JSON
```
中读取关键统计数据：
- total_sessions
- total_messages
- total_duration_hours
- date_range
从
```
$INSIGHTS_JSON
```
中读取执行摘要
向用户展示：

示例输出：

✅ 洞察报告已生成

📊 摘要：
- 分析的会话数：42
- 总消息数：1,247
- 时间范围：2026-01-15 至 2026-02-06
- 总时长：18.5小时

🎯 执行摘要：
- 有效做法：您能有效利用Claude进行快速原型开发...
- 待改进点：切换项目时频繁重置上下文...
- 快速优化建议：尝试使用/compact减少令牌消耗...
- 进阶工作流：探索使用LSP工具进行多文件重构...

📄 完整报告：file://$REPORT_PATH

在浏览器中打开报告，查看8个类别的详细洞察。

Output Format

输出格式

The generated HTML report includes:

Header:

Title: "Insights Report"
CLI type badge
Generation date
Stats bar (sessions · messages · hours · commits)

8 Sections:

At a Glance - 4-item executive summary
What You Work On - Project areas with session counts
Your Style - Interaction style narrative
What's Working Well - Impressive workflows
Friction Points - Categorized pain points
Suggestions - Features and usage patterns to try
On the Horizon - Future opportunities with copyable prompts
Fun Moment - Memorable headline

Features:

Self-contained (no external resources)
Dark/light theme support (auto-detects via
```
prefers-color-scheme
```
)
Responsive design (mobile-friendly)
Print-friendly styles

生成的HTML报告包含以下内容：

页眉：

标题："Insights Report"
CLI类型标识
生成日期
统计栏（会话数 · 消息数 · 时长 · 提交数）

8个板块：

执行摘要 - 4项内容的高管摘要
您的工作领域 - 带有会话数的项目领域
您的使用风格 - 交互风格叙述
有效做法 - 高效工作流
痛点 - 分类后的问题点
建议 - 可尝试的功能和使用模式
未来展望 - 包含可复制提示词的未来机会
难忘时刻 - 引人注目的标题

特性：

独立运行（无外部资源）
支持深色/浅色主题（通过
```
prefers-color-scheme
```
自动检测）
响应式设计（适配移动端）
适合打印的样式

Reference Files

参考文件

All supporting documentation is in

references/

analysis-prompts.md - Complete text of all 8 analysis prompts with expected JSON schemas
facet-extraction.md - Session facet extraction prompt and enums
schema.md - JSON schemas for session metadata, stats, facets, and insights
cli-formats.md - Detailed documentation of session storage formats for each CLI
suggestions-by-cli.md - CLI-specific feature suggestions (MCP servers, custom skills, hooks, etc.)

所有支持文档都在

references/

目录中：

analysis-prompts.md - 所有8个分析提示词的完整文本及预期JSON schema
facet-extraction.md - 会话维度提取提示词及枚举值
schema.md - 会话元数据、统计数据、维度和洞察的JSON schema
cli-formats.md - 各CLI的会话存储格式详细文档
suggestions-by-cli.md - 特定于CLI的功能建议（MCP服务器、自定义Skill、钩子等）

Troubleshooting

故障排除

"jq: command not found"

Solution: Install jq:

macOS:
```
brew install jq
```
Linux:
```
apt install jq
```
or
```
yum install jq
```

解决方案： 安装jq：

macOS：
```
brew install jq
```
Linux：
```
apt install jq
```
或
```
yum install jq
```

"Bash version too old"

Solution: Upgrade to bash 4+:

macOS:
```
brew install bash
```
(then restart terminal)
Linux: Usually already 4+

解决方案： 升级到bash 4+：

macOS：
```
brew install bash
```
（然后重启终端）
Linux：通常已预装4+版本

"No sessions found"

Possible causes:

CLI not detected correctly

Run:
```
./scripts/detect-cli.sh
```

Override:

./scripts/detect-cli.sh --cli claude-code

Session directory empty

Check:
```
ls ~/.claude/projects/
```
(Claude Code)

Check:

ls ~/.local/share/opencode/storage/session/

(OpenCode)

Check:
```
ls ~/.codex/sessions/
```
(Codex)

Sessions filtered out (<2 messages, <1 minute)
- Review filtering logic in
```
scripts/collect-sessions.sh
```

可能原因：

CLI检测错误

执行：
```
./scripts/detect-cli.sh
```

手动覆盖：

./scripts/detect-cli.sh --cli claude-code

会话目录为空
- 检查：
```
ls ~/.claude/projects/
```
  （Claude Code）
- 检查：
```
ls ~/.local/share/opencode/storage/session/
```
  （OpenCode）
- 检查：
```
ls ~/.codex/sessions/
```
  （Codex）
会话被过滤掉（消息数<2，时长<1分钟）
- 查看
```
scripts/collect-sessions.sh
```
  中的过滤逻辑

"Report generation fails"

Possible causes:

Invalid JSON from analysis prompts
- Verify:
```
echo "$INSIGHTS_JSON" | jq .
```
- Check: Each prompt response is valid JSON
Missing template file
- Verify:
```
ls scripts/templates/report.html
```
Python not available
- generate-report.sh uses Python for JSON injection
- Verify:
```
python3 --version
```

可能原因：

分析提示词返回无效JSON
- 验证：
```
echo "$INSIGHTS_JSON" | jq .
```
- 检查：每个提示词的响应是否为有效JSON
缺少模板文件
- 验证：
```
ls scripts/templates/report.html
```
Python不可用
- generate-report.sh使用Python注入JSON
- 验证：
```
python3 --version
```

"Analysis prompts return non-JSON"

Solution: Ensure each prompt includes "RESPOND WITH ONLY A VALID JSON OBJECT" instruction. If the model returns markdown code blocks, strip them:

bash

response=$(echo "$response" | sed 's/^```json//; s/^```//; s/```$//')

解决方案： 确保每个提示词都包含“仅返回有效的JSON对象”的指令。如果模型返回markdown代码块，请移除：

bash

response=$(echo "$response" | sed 's/^```json//; s/^```//; s/```$//')

Examples

示例

Basic Usage

基本用法

bash

cd skills/insights

bash

cd skills/insights

1. Check dependencies

1. 检查依赖项

./scripts/check-deps.sh || exit 1

2. Detect CLI

2. 检测CLI

CLI_TYPE=$(./scripts/detect-cli.sh)

3. Collect sessions

3. 收集会话

SESSIONS=$(./scripts/collect-sessions.sh --cli "$CLI_TYPE" --limit 50)

4. Aggregate stats

4. 聚合统计数据

STATS=$(echo "$SESSIONS" | ./scripts/aggregate-stats.sh)

5. Extract facets (agent logic, not shown)

5. 提取维度（Agent逻辑，未展示）

6. Run analysis prompts (agent logic, not shown)

6. 运行分析提示词（Agent逻辑，未展示）

7. Generate report

7. 生成报告

STATS_FILE=$(mktemp) echo "$STATS" > "$STATS_FILE" REPORT=$(echo "$INSIGHTS" | ./scripts/generate-report.sh --stats "$STATS_FILE" --cli "$CLI_TYPE")

8. Present

8. 展示结果

echo "Report: file://$REPORT"

undefined

echo "Report: file://$REPORT"

undefined

Override CLI Detection

手动覆盖CLI检测

bash

undefined

bash

undefined

Force Claude Code

强制使用Claude Code

CLI_TYPE=$(./scripts/detect-cli.sh --cli claude-code)

undefined

CLI_TYPE=$(./scripts/detect-cli.sh --cli claude-code)

undefined

Limit Sessions

限制会话数量

bash

undefined

bash

undefined

Analyze only last 10 sessions

仅分析最近10个会话

SESSIONS=$(./scripts/collect-sessions.sh --cli "$CLI_TYPE" --limit 10)

undefined

SESSIONS=$(./scripts/collect-sessions.sh --cli "$CLI_TYPE" --limit 10)

undefined

Custom Output Path

自定义输出路径

bash

undefined

bash

undefined

Save to specific location

保存到指定位置

mkdir -p ~/reports REPORT=$(echo "$INSIGHTS" | ./scripts/generate-report.sh
--stats stats.json
--cli claude-code
--output ~/reports/insights-$(date +%Y%m%d).html)

undefined

mkdir -p ~/reports REPORT=$(echo "$INSIGHTS" | ./scripts/generate-report.sh
--stats stats.json
--cli claude-code
--output ~/reports/insights-$(date +%Y%m%d).html)

undefined

Notes

注意事项

Token cost: Facet extraction (Step 5) and analysis prompts (Step 6) consume tokens. Limit facet extraction to 20 sessions.
Caching: Cache extracted facets in
```
~/.agent-insights/cache/facets/
```
to avoid re-extraction on subsequent runs.
Parallelization: Run the 8 analysis prompts in parallel for faster execution.
Privacy: All data stays local. No external API calls. Report is self-contained HTML.
Extensibility: To add a new CLI, create an adapter in
```
collect-sessions.sh
```
following the existing pattern.

令牌成本： 维度提取（步骤5）和分析提示词（步骤6）会消耗令牌。请将维度提取限制为20个会话。
缓存： 将提取的维度缓存到
```
~/.agent-insights/cache/facets/
```
中，避免后续运行时重复提取。
并行化： 并行运行8个分析提示词以加快执行速度。
隐私： 所有数据都保存在本地，无外部API调用。报告是独立的HTML文件。
可扩展性： 要添加新的CLI，请按照现有模式在
```
collect-sessions.sh
```
中创建适配器。

Language Support

语言支持

The insights skill supports multi-language report generation.

Insights Skill支持多语言报告生成。

Supported Languages

支持的语言

Built-in: English (
```
en
```
), Korean (
```
ko
```
)
Dynamic: Any language via AI translation

内置： 英语（
```
en
```
）、韩语（
```
ko
```
）
动态： 通过AI翻译支持任意语言

How to Specify Language

指定语言的方式

CLI flag:
```
--language <code>
```
(e.g.,
```
--language ko
```
)
Environment variable:
```
LANG
```
or
```
LC_ALL
```
(e.g.,
```
LANG=ko_KR.UTF-8
```
)
Default: English (
```
en
```
)

CLI参数：
```
--language <code>
```
（例如：
```
--language ko
```
）
环境变量：
```
LANG
```
或
```
LC_ALL
```
（例如：
```
LANG=ko_KR.UTF-8
```
）
默认值： 英语（
```
en
```
）

What Gets Translated

翻译内容

HTML report UI strings: Section titles, labels (26 strings)
Analysis content: Insights JSON values (via prompt instruction)

HTML报告UI字符串： 板块标题、标签（26个字符串）
分析内容： 洞察JSON值（通过提示词指令）

What Stays in English

保留英文的内容

JSON keys: Always English for consistency
Shell script output: Help messages, errors

JSON键： 始终使用英文以保持一致性
Shell脚本输出： 帮助信息、错误信息

Version

版本

This skill targets:

Claude Code 2.1+
OpenCode 1.1+
Codex (latest)

For older versions, session formats may differ. Check

references/cli-formats.md

for compatibility notes.

此Skill针对以下版本：

Claude Code 2.1+
OpenCode 1.1+
Codex（最新版）

对于旧版本，会话格式可能不同。请查看

references/cli-formats.md

中的兼容性说明。