← Back to Skills Marketplace
legionspace-hackathon

AI Coach Batch Session Summary

by LegionSpace-Hackathon · GitHub ↗ · v1.1.0 · MIT-0
cross-platform ✓ Security Clean
142
Downloads
0
Stars
0
Active Installs
2
Versions
Install in OpenClaw
/install ai-coach-batch-session-summary
Description
录音转写洞察:先 parse-user-intent 反问确认时间与维度,再 fetch 拉数;默认近 1 个月 + 十维;默认报告页面。 触发:对象词(含教练/coach)+ 动作词,或完整口令,或显式 intent=ai-coach-batch-session-summary。
README (SKILL.md)

AI Coach Batch Session Summary

目标

拉取 legion-hardware 的 ASR 转写并生成报告。仅依据 asrText,禁止使用 aiJiaolianResultJson / aiSummaryContent 作为结论依据。


工作流程(必须按序,禁止跳步)

1. parse-user-intent.mjs     → 解析时间 / 维度 / 交付形态
2. needClarification=true    → 反问用户,结束(禁止执行 fetch)
3. 用户补充或 skipClarification / confirmed* → 再执行 fetch-asr-recordings.mjs
4. 按 reportPlan + themeRelated + outputFormat 生成报告

禁止只跑 fetch-asr-recordings.mjs 而不先意图解析(fetch 内置相同解析,若需反问会直接 ok:false 且不拉数)。

脚本调用

# 步骤 1:意图(不拉数)
echo '{"user":"u","model":"openclaw:u123","userMessage":"帮我总结录音"}' | \
  node "{baseDir}/scripts/parse-user-intent.mjs"

# 步骤 2:确认后拉数(token 仅来自 Header / LEGION_AUTH_TOKEN)
export LEGION_AUTH_TOKEN="\x3CJWT>"
echo '{"user":"u","model":"openclaw:u123","userMessage":"总结最近 5 天录音","skipClarification":true}' | \
  node "{baseDir}/scripts/fetch-asr-recordings.mjs"

默认值(反问后仍不明确,或用户说「按默认」)

默认 报告说明
时间 1 个月 「未确认具体时间,已按默认最近 1 个月统计」
维度 十维拓客 「因未提取到明确的分析维度信息,已使用默认十维拓客统计框架处理」

反问沟通

parse-user-intent / fetch 在以下情况 needClarification: true

信号 反问要点
timeRange.confidence = ambiguous 几天 / 几周 / 几个月
timeRange.confidence = missing 统计窗口;不答则默认 1 个月
analysisFocus.confidence = ambiguous 拓客 / 话术 / 教练 / 客户需求 / 或默认十维
analysisFocus.confidence = missing 分析方向;不答则默认十维

跳过反问:

  • 用户话术中「就按默认 / 你看着办 / 用默认」
  • body:skipClarification: true(网关在用户回复后设置)
  • body:confirmedTime / confirmedDimension(见下)

多轮合并 userMessage

网关将 首轮 + 追问回复 写入同一 body:

{
  "userMessage": "帮我总结录音",
  "followUpMessage": "最近 5 天,按拓客复盘",
  "skipClarification": true
}

或拼接为单字段 userMessage。脚本会合并 userMessage + query + followUpMessage

网关确认字段(可选)

字段 示例 作用
skipClarification true 不再反问,按解析结果或默认拉数
confirmedTime "default"{ "startTime":"...","endTime":"..." } 锁定时间窗
confirmedDimension "default_ten" / "ten" / "AI 教练" / "股票趋势" 锁定维度或专题

触发词速查

激活(其一): 对象词 + 动作词;完整口令intent=ai-coach-batch-session-summary

动作词: 总结、汇总、归纳、梳理、概括、小结、复盘、盘点、整合、整理、摘要、报告、结论(及英文同义,须配对象词)。

对象词:

类别 中文 英文
录音 / 转写 录音、转写、ASR recording, transcript, ASR
拓客 拓客、走访、地推 outreach, field sales
话术 / 客户 话术、客户拜访 pitch, script, customer visit
教练 教练、AI 教练、教练反馈、教练点评 coach, coaching, AI coach, coach feedback
系统 legion-hardware legion, hardware

完整口令

  • 总结最近 5 天录音
  • 十维拓客报告
  • 总结一下教练反馈
  • AI coach recap for last month

不触发: 仅「总结一下」无对象词;调用 POST /api/recordings/ai-jiaolian 且无话术总结类动作词(走 AI 教练单次能力,不走本 skill)。


入站参数(OpenClaw / 网关)

{
  "user": "OpenClaw 用户标识(必填,映射为 hardware userId)",
  "model": "openclaw:{user}{orgId} 或 openclaw:{user}person",
  "userMessage": "用户原话(必填,可在 metadata)",
  "followUpMessage": "可选,第二轮补充",
  "skipClarification": false,
  "confirmedTime": "default",
  "confirmedDimension": "default_ten",
  "orgId": "可选;网关已从 model 解析时可直传"
}

Header:Authorization: Bearer {JWT}禁止 body.token)

userId / orgId 取值(resolve-openclaw-identity.mjs

与 tfd-api AiService 的 model 组装一致:

model 形态 含义 hardware 调用
openclaw:{user}{orgId} 组织上下文 userId=userorgId 为去掉前缀 openclaw:{user} 后的后缀
openclaw:{user}person 个人上下文 userId=user不传 orgId(或为空)

规则:

  1. userId = 请求体字段 user(兼容 userId)。
  2. model 以 person 结尾(不区分大小写)→ 个人模式,不传 orgId
  3. model 非 personorgId = model 去掉前缀 openclaw: + user 后的剩余部分。
  4. 若 body 已带非空 orgId(如网关 metadata)→ 优先使用,不再从 model 解析。
  5. 若 body 显式 orgId: "" → 视为个人模式。

示例(user=abc):

model orgId(查询)
openclaw:abc123 123
openclaw:abcperson 不传(个人)
openclaw:abc 不传(后缀为空)

时间规则(parse-time-range.mjs

表述 confidence 行为
最近 5 天 / 近 2 个月 explicit 按解析查询
最近 / 近期(无数字) ambiguous 先反问
无时间 missing 先反问
反问后仍无 default 1 个月

时区 Asia/Shanghai;上限:月 ≤12,天 ≤366。


维度与报告分支

条件 输出
无显式专题(走十维) 十维拓客报告 + 默认维度说明(若适用)
用户明确「十维 / 默认十维」 十维(无需「未提取维度」说明)
显式专题且 themeRelated=true ## 用户关注:{主题}
显式专题且 themeRelated=false ## 分析结果无相关内容(禁止十维、禁止编造)

themeRelated 由脚本用 主题核心词 匹配 asrText(已避免单独「KYC」「趋势」误命中)。

样本量

recordCount 处理
0 十维骨架或专题内写明「样本为空」
\x3C 5 标注「小样本,结论仅供参考」
> 80(默认) 见 stdout recordsHint,分析时优先近期并抽样

数据拉取与部署

说明
默认 baseUrl http://192.168.109.50:8900LEGION_HARDWARE_BASE_URL 可覆盖)
接口 GETPOST /api/recordings/asr-completed + startTime/endTime
部署 新 jar(POST 支持 body 内时间);旧版 POST 仅 3 个月固定窗,自定义时间会失效

fetch 成功时 stdout 含:queryanalysisFocusthemeRelatedreportPlanoutputFormatrecords[]


交付形态(outputFormat

mode 行为
page(默认) templates/insight-report-page.md 生成完整报告页(Markdown/HTML)
text 用户要求「只要文字/不要页面」
custom 用户要求导出/邮件等,从其要求

十维拓客(默认维度)

  1. 拓客场景 2. 触达人群 3. 开口获客话术 4. 客户需求 5. 客户拒绝原因
  2. 有效意向分类 7. 产品匹配 8. 成交阻碍痛点 9. 跟进动作 10. 拓客成效数据

每维:关键发现 / 典型摘录 / 量化指标 / 优化建议。


错误与安全

  • user / Authorization → 不调 hardware;个人模式可无 orgId
  • needClarification 时禁止编造数据
  • 勿暴露 token、内网 IP/端口
  • user / 解析出的 orgId 宜与业务身份一致(建议网关校验)

检查清单

  • 已跑 parse-user-intentneedClarification 时已反问且未 fetch
  • 默认时间/维度说明已写入报告(若适用)
  • themeUnrelated 时仅输出「无相关内容」
  • 交付形态符合 outputFormat
  • 未使用 body.token
  • user / model 已按规则解析;person 模式未传 orgId
Usage Guidance
Treat this as an incomplete review, not a clean security endorsement. Re-run the scan when metadata.json and the artifact directory can be read.
Capability Assessment
Purpose & Capability
Artifact review could not be completed because local command execution failed before metadata.json or artifact files could be read.
Instruction Scope
No SKILL.md instructions were available for evidence-backed scope assessment.
Install Mechanism
No install metadata or manifest contents were available for evidence-backed install assessment.
Credentials
No artifact contents were available to assess requested environment access against purpose.
Persistence & Privilege
No artifact contents were available to assess persistence, credential, or privilege behavior.
How to Use
  1. Make sure OpenClaw is installed (local or Docker)
  2. Run the install command in chat: /install ai-coach-batch-session-summary
  3. After installation, invoke the skill by name or use /ai-coach-batch-session-summary
  4. Provide required inputs per the skill's parameter spec and get structured output
Version History
v1.1.0
Add resolve-openclaw-identity.mjs for automatic userId/orgId resolution from model field
v1.0.0
Initial release: AI 教练批量会话总结,聚焦教练反馈/场景对练/教练点评的汇总分析
Metadata
Slug ai-coach-batch-session-summary
Version 1.1.0
License MIT-0
All-time Installs 0
Active Installs 0
Total Versions 2
Frequently Asked Questions

What is AI Coach Batch Session Summary?

录音转写洞察:先 parse-user-intent 反问确认时间与维度,再 fetch 拉数;默认近 1 个月 + 十维;默认报告页面。 触发:对象词(含教练/coach)+ 动作词,或完整口令,或显式 intent=ai-coach-batch-session-summary。 It is an AI Agent Skill for Claude Code / OpenClaw, with 142 downloads so far.

How do I install AI Coach Batch Session Summary?

Run "/install ai-coach-batch-session-summary" in the OpenClaw or Claude Code chat to install it in one step — no extra setup required.

Is AI Coach Batch Session Summary free?

Yes, AI Coach Batch Session Summary is completely free, licensed under MIT-0. You can download, install and use it at no cost.

Which platforms does AI Coach Batch Session Summary support?

AI Coach Batch Session Summary is cross-platform and runs anywhere OpenClaw / Claude Code is available (cross-platform).

Who created AI Coach Batch Session Summary?

It is built and maintained by LegionSpace-Hackathon (@legionspace-hackathon); the current version is v1.1.0.

💬 Comments