Interview Minutes Analysis
Use this when someone asks to analyze pending Lark Minutes interview records and notify HR.
You (the agent) run the WHOLE flow yourself with your own tools, then WRITE the results to Lark
Bitable and post a summary to the chat. Do NOT refuse, do NOT just describe, do NOT claim "the
skill only formats" — drive the flow. If a data source or the target base is unavailable, SKIP that
part and still post what you have. Never ask the user for tokens — your NyxID-brokered tools broker
all credentials.
How to run it
Resolve inputs and target. Use the flow's ATS Bitable and HR recipient unless the user
overrides them: target Bitable app_token FSl0bCi9raBuLbsdTbHlgb0agwf, table_id
tblgZgSqmeBag2na; HR open_id ou_4473a6a9feeca68d2eeaf2cb95416f37. If the user named a
specific record or Minutes URL, use it; otherwise process the first pending ATS record. If the
ATS base is unresolvable, proceed in summary-only mode.
COLLECT the ATS record. Read the ATS Bitable via nyxid_proxy {slug:"api-lark-bot", path:"/open-apis/bitable/v1/apps/<app_token>/tables/<table_id>/records?page_size=100&filter=<urlencode('CurrentValue.[Minutes URL]!=\"\"')>", method:"GET"} (or the equivalent lark_* tool). Pick the first item whose Minutes URL
contains /minutes/ and whose AI Interview Notes is empty. Keep record_id, Minutes URL,
Candidate Name, Job Title, Job Description, and Interviewer. If there is no pending
record, post that and stop.
COLLECT the Minutes transcript. Extract minute_token with /minutes/([a-z0-9]+)/i, then
read Minutes via nyxid_proxy slug api-lark-bot (or lark_minutes):
/open-apis/minutes/v1/minutes/{minute_token} and
/open-apis/minutes/v1/minutes/{minute_token}/transcript?need_speaker=true&need_timestamp=true&file_format=txt.
Use the returned title, else <candidate_name> 面试. If the transcript is empty, post that it may
still be processing and stop.
ANALYZE. With your own model, classify the role as 技术岗, 商业/职能岗, or 家政/个人助理岗
from the title/JD/transcript, run the role-specific veto checks, and return valid JSON with
veto_check, overall, score, summary, strengths, red_flags, key_moments,
culture_fit, culture_fit_reason, next_step, suggested_questions, and veto_count.
- 商业/职能岗: AI 实质改造、具体 AI 案例、独立贡献、从 0 到 1、思维深度。
- 技术岗: AI 编程实质、Agent/AI 技术栈、系统设计能力、独立交付能力,
v5 填 N/A。
- 家政/个人助理岗: 诚信核查、稳定性、核心技能,
v4 和 v5 填 N/A。
Any VETO caps score at 2; multiple VETOs normally mean score: 1 and overall: "Strong Pass".
Base every judgment on transcript evidence.
BUILD the exact payloads. Run scripts/build_interview_analysis_payload.js via code_execute
(language "javascript"; NO stdin — inline the data into the code):
console.log(JSON.stringify(buildPayload({ mode: "build_payload", analysis, hrOpenId, atsAppToken, atsTableId, recordId, generatedAt, candidateName, jobTitle, minuteTitle }))), where generatedAt
is the current Asia/Singapore timestamp. Use the output's lark.bitableUpdate.body and
lark.cardMessage.body. If code execution is unavailable, follow
references/interview-analysis-contract.md field-for-field.
WRITE the notes, then post the card. Update the ATS record via nyxid_proxy
{slug:"api-lark-bot", path:"/open-apis/bitable/v1/apps/<app_token>/tables/<table_id>/records/<record_id>", method:"PUT", body:<lark.bitableUpdate.body>}. THEN send the HR interactive card with lark_messages_send (or
reply_with_interaction); the equivalent is nyxid_proxy
POST /open-apis/im/v1/messages?receive_id_type=open_id with lark.cardMessage.body. If no ATS
target was resolved OR the update fails, SKIP the Bitable write and STILL post the analysis card /
summary to the chat — never fail the run for a missing base.
REPORT one line: candidate, job title, score/overall, ATS record updated (or "summary-only"),
HR card sent.
Payload builder reference
Bundled script: scripts/build_interview_analysis_payload.js. Accepts mode: "build_payload" (or
omitted); other modes return skip: true. Required after alias normalization: analysis,
hrOpenId, atsAppToken, atsTableId, recordId, generatedAt. Optional: candidateName,
jobTitle, minuteTitle. analysis may be a direct object or a Groq-style
choices[0].message.content string. Output includes notes_text, card, lark_card_json,
lark.bitableUpdate (PUT), and lark.cardMessage (POST to open_id). Exact aliases, fallback
parsing, card fields, and request bodies are in references/interview-analysis-contract.md.
Guardrails
- Only use the Lark records, Minutes metadata, transcript text, and analysis evidence the tools
actually returned — never invent ids, records, statuses, quotes, or candidate details.
- Prefer running the bundled script; it is the deterministic source of truth for payload shape.
- All Lark access, sends, and credential exchange go through your NyxID-brokered tools or Lark
tools; never handle raw secrets.
1---2name: interview-minutes-analysis-payload-builder3version: "2.1"4description: Runs the interview-minutes analysis flow end to end — scans the ATS Bitable, reads the Lark Minutes transcript, analyzes the interview with your own model, writes the AI notes back to the ATS record, and posts the HR card.5metadata:6 source: https://skillmd.com/skills/chronoaiproject/interview-minutes-analysis-payload-builder7 category: plain8 tag:9 - aevatar10 - s-capability11 - interview12 - lark13 - payload-builder14 clawdbot:15 emoji: "memo"16 files:17 - "references/*"18 - "scripts/*"19---2021# Interview Minutes Analysis2223Use this when someone asks to analyze pending Lark Minutes interview records and notify HR.2425**You (the agent) run the WHOLE flow yourself with your own tools, then WRITE the results to Lark26Bitable and post a summary to the chat.** Do NOT refuse, do NOT just describe, do NOT claim "the27skill only formats" — drive the flow. If a data source or the target base is unavailable, SKIP that28part and still post what you have. Never ask the user for tokens — your NyxID-brokered tools broker29all credentials.3031## How to run it32331. **Resolve inputs and target.** Use the flow's ATS Bitable and HR recipient unless the user34 overrides them: target Bitable `app_token` `FSl0bCi9raBuLbsdTbHlgb0agwf`, `table_id`35 `tblgZgSqmeBag2na`; HR `open_id` `ou_4473a6a9feeca68d2eeaf2cb95416f37`. If the user named a36 specific record or Minutes URL, use it; otherwise process the first pending ATS record. If the37 ATS base is unresolvable, proceed in summary-only mode.38392. **COLLECT the ATS record.** Read the ATS Bitable via `nyxid_proxy` `{slug:"api-lark-bot",40 path:"/open-apis/bitable/v1/apps/<app_token>/tables/<table_id>/records?page_size=100&filter=<urlencode('CurrentValue.[Minutes URL]!=\"\"')>",41 method:"GET"}` (or the equivalent `lark_*` tool). Pick the first item whose `Minutes URL`42 contains `/minutes/` and whose `AI Interview Notes` is empty. Keep `record_id`, `Minutes URL`,43 `Candidate Name`, `Job Title`, `Job Description`, and `Interviewer`. If there is no pending44 record, post that and stop.45463. **COLLECT the Minutes transcript.** Extract `minute_token` with `/minutes/([a-z0-9]+)/i`, then47 read Minutes via `nyxid_proxy` slug `api-lark-bot` (or `lark_minutes`):48 `/open-apis/minutes/v1/minutes/{minute_token}` and49 `/open-apis/minutes/v1/minutes/{minute_token}/transcript?need_speaker=true&need_timestamp=true&file_format=txt`.50 Use the returned title, else `<candidate_name> 面试`. If the transcript is empty, post that it may51 still be processing and stop.52534. **ANALYZE.** With your own model, classify the role as `技术岗`, `商业/职能岗`, or `家政/个人助理岗`54 from the title/JD/transcript, run the role-specific veto checks, and return valid JSON with55 `veto_check`, `overall`, `score`, `summary`, `strengths`, `red_flags`, `key_moments`,56 `culture_fit`, `culture_fit_reason`, `next_step`, `suggested_questions`, and `veto_count`.57 - 商业/职能岗: AI 实质改造、具体 AI 案例、独立贡献、从 0 到 1、思维深度。58 - 技术岗: AI 编程实质、Agent/AI 技术栈、系统设计能力、独立交付能力,`v5` 填 `N/A`。59 - 家政/个人助理岗: 诚信核查、稳定性、核心技能,`v4` 和 `v5` 填 `N/A`。60 Any VETO caps `score` at 2; multiple VETOs normally mean `score: 1` and `overall: "Strong Pass"`.61 Base every judgment on transcript evidence.62635. **BUILD the exact payloads.** Run `scripts/build_interview_analysis_payload.js` via `code_execute`64 (language `"javascript"`; NO stdin — inline the data into the code):65 `console.log(JSON.stringify(buildPayload({ mode: "build_payload", analysis, hrOpenId, atsAppToken,66 atsTableId, recordId, generatedAt, candidateName, jobTitle, minuteTitle })))`, where `generatedAt`67 is the current Asia/Singapore timestamp. Use the output's `lark.bitableUpdate.body` and68 `lark.cardMessage.body`. If code execution is unavailable, follow69 `references/interview-analysis-contract.md` field-for-field.70716. **WRITE the notes, then post the card.** Update the ATS record via `nyxid_proxy`72 `{slug:"api-lark-bot",73 path:"/open-apis/bitable/v1/apps/<app_token>/tables/<table_id>/records/<record_id>", method:"PUT",74 body:<lark.bitableUpdate.body>}`. THEN send the HR interactive card with `lark_messages_send` (or75 `reply_with_interaction`); the equivalent is `nyxid_proxy`76 `POST /open-apis/im/v1/messages?receive_id_type=open_id` with `lark.cardMessage.body`. **If no ATS77 target was resolved OR the update fails, SKIP the Bitable write and STILL post the analysis card /78 summary to the chat — never fail the run for a missing base.**79807. **REPORT** one line: candidate, job title, score/overall, ATS record updated (or "summary-only"),81 HR card sent.8283## Payload builder reference8485Bundled script: `scripts/build_interview_analysis_payload.js`. Accepts `mode: "build_payload"` (or86omitted); other modes return `skip: true`. Required after alias normalization: `analysis`,87`hrOpenId`, `atsAppToken`, `atsTableId`, `recordId`, `generatedAt`. Optional: `candidateName`,88`jobTitle`, `minuteTitle`. `analysis` may be a direct object or a Groq-style89`choices[0].message.content` string. Output includes `notes_text`, `card`, `lark_card_json`,90`lark.bitableUpdate` (PUT), and `lark.cardMessage` (POST to `open_id`). Exact aliases, fallback91parsing, card fields, and request bodies are in `references/interview-analysis-contract.md`.9293## Guardrails9495- Only use the Lark records, Minutes metadata, transcript text, and analysis evidence the tools96 actually returned — never invent ids, records, statuses, quotes, or candidate details.97- Prefer running the bundled script; it is the deterministic source of truth for payload shape.98- All Lark access, sends, and credential exchange go through your NyxID-brokered tools or Lark99 tools; never handle raw secrets.