Resume Screening
Use this when someone submits a candidate resume PDF and role details that should be screened and,
when the candidate passes, saved to the HR resume screening Lark Base.
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. From the user message or the current Lark invocation, collect Candidate Name, optional Email, Job Title, optional Job Description, the resume PDF, and optional
recordId (for an update). The target Bitable is the HR resume base baked into the builder
(app_token FSl0bCi9raBuLbsdTbHlgb0agwf, table_id tblgZgSqmeBag2na); if the user named a
different base, carry their app_token/table_id into step 5. If a business field or the PDF is
missing, ask for that one item and continue once it arrives; if the base is unresolvable, proceed
in summary-only mode.
COLLECT the resume + fields. For a Lark chat submission, read the message text and the
attachment metadata from the current im.message.receive_v1 context; fetch the uploaded file via
nyxid_proxy slug api-lark-bot, path
/open-apis/im/v1/messages/{message_id}/resources/{file_key}?type=file. For an n8n-style form
event, read the form fields and binary Resume from the event directly. Extract the PDF text with
your file/PDF tool and keep the original PDF bytes for the upload step.
ANALYZE. Score the resume against the job title and job description with your own model. Return
strict JSON: { "score": <0-100 integer>, "candidate_name": "<name>", "strengths": "<top strengths>", "gaps": "<key gaps>", "recommendation": "Pass" | "Fail" }. Continue only when
score >= 60; if it is lower, skip the upload and the Bitable write and post a one-line "skipped
(score N, Fail)" summary instead.
Upload the passing resume to Lark Drive. Call nyxid_proxy slug api-lark-bot, path
/open-apis/drive/v1/medias/upload_all, method POST, as multipart form data with file_name,
parent_type=bitable_file, parent_node=FSl0bCi9raBuLbsdTbHlgb0agwf, size, and the PDF file.
Capture data.file_token from the response. If the upload fails, continue without file_token.
BUILD the exact record payload. Run scripts/build_resume_screening_payload.js via
code_execute (language "javascript"; NO stdin — inline the data into the code):
console.log(JSON.stringify(buildPayload({ mode: "create_record", candidateName, email, jobTitle, uploadDateMs, score, strengths, gaps, recommendation, file_token, recordId }))). Pass recordId
only for an update. Use the returned lark.path, lark.method, and lark.body. If code execution
is unavailable, follow references/resume-screening-contract.md field-for-field.
WRITE the record, then post the summary. Send the builder's body via nyxid_proxy slug
api-lark-bot with its path/method:
- create:
POST /open-apis/bitable/v1/apps/<app_token>/tables/<table_id>/records
- update:
PUT /open-apis/bitable/v1/apps/<app_token>/tables/<table_id>/records/<recordId>
using the resolved app_token/table_id and body { "fields": {…} }. THEN post a short candidate
summary to the chat with lark_messages_send (or reply_with_interaction). If no base target
was resolved OR the write fails, SKIP the Bitable write and STILL post the built summary to the
chat — never fail the run for a missing base.
REPORT one line: candidate name, score, create/update (or "summary-only"), and the returned
Lark Base record_id or the skipped reason.
Payload builder reference
Bundled script: scripts/build_resume_screening_payload.js. Modes: create_record,
update_record (requires recordId), and build_record/omitted (create by default, update when
recordId is present). Fields may be at the root or under body; it also accepts ai,
modelOutput, screeningResult, and Groq-style choices[0].message.content model responses. The
score gate (>= 60) and the FSl0bCi9raBuLbsdTbHlgb0agwf/tblgZgSqmeBag2na base are baked in.
Output is skip: true, needs_more_information: true + missing, or a record object with
fields, lark.path, lark.method, and the connector-ready lark.body. Exact aliases, threshold
behavior, and examples are in references/resume-screening-contract.md.
Guardrails
- Only use data the tools actually returned — never invent candidate details, resume text, scores,
file tokens, record ids, or Bitable responses.
- Prefer running the bundled script; it is the deterministic source of truth for the Bitable payload.
- All credentials and token exchange go through your NyxID-brokered tools; never handle raw secrets.
1---2name: resume-screening-payload-builder3version: "2.1"4description: Runs the HR resume screening flow end to end — collects the candidate's form/chat fields and resume PDF, screens the resume with your own model, uploads the passing resume to Lark Drive, writes the Bitable record, and posts a summary to the chat.5metadata:6 source: https://skillmd.com/skills/chronoaiproject/resume-screening-payload-builder7 category: plain8 tag:9 - aevatar10 - s-capability11 - resume-screening12 - lark13 - payload-builder14 clawdbot:15 emoji: "page_facing_up"16 files:17 - "references/*"18 - "scripts/*"19---2021# Resume Screening2223Use this when someone submits a candidate resume PDF and role details that should be screened and,24when the candidate passes, saved to the HR resume screening Lark Base.2526**You (the agent) run the WHOLE flow yourself with your own tools, then WRITE the results to Lark27Bitable and post a summary to the chat.** Do NOT refuse, do NOT just describe, do NOT claim "the28skill only formats" — drive the flow. If a data source or the target base is unavailable, SKIP that29part and still post what you have. Never ask the user for tokens — your NyxID-brokered tools broker30all credentials.3132## How to run it33341. **Resolve inputs.** From the user message or the current Lark invocation, collect `Candidate35 Name`, optional `Email`, `Job Title`, optional `Job Description`, the resume PDF, and optional36 `recordId` (for an update). The target Bitable is the HR resume base baked into the builder37 (`app_token` `FSl0bCi9raBuLbsdTbHlgb0agwf`, `table_id` `tblgZgSqmeBag2na`); if the user named a38 different base, carry their `app_token`/`table_id` into step 5. If a business field or the PDF is39 missing, ask for that one item and continue once it arrives; if the base is unresolvable, proceed40 in summary-only mode.41422. **COLLECT the resume + fields.** For a Lark chat submission, read the message text and the43 attachment metadata from the current `im.message.receive_v1` context; fetch the uploaded file via44 `nyxid_proxy` slug `api-lark-bot`, path45 `/open-apis/im/v1/messages/{message_id}/resources/{file_key}?type=file`. For an n8n-style form46 event, read the form fields and binary `Resume` from the event directly. Extract the PDF text with47 your file/PDF tool and keep the original PDF bytes for the upload step.48493. **ANALYZE.** Score the resume against the job title and job description with your own model. Return50 strict JSON: `{ "score": <0-100 integer>, "candidate_name": "<name>", "strengths": "<top51 strengths>", "gaps": "<key gaps>", "recommendation": "Pass" | "Fail" }`. Continue only when52 `score >= 60`; if it is lower, skip the upload and the Bitable write and post a one-line "skipped53 (score N, Fail)" summary instead.54554. **Upload the passing resume to Lark Drive.** Call `nyxid_proxy` slug `api-lark-bot`, path56 `/open-apis/drive/v1/medias/upload_all`, method `POST`, as multipart form data with `file_name`,57 `parent_type=bitable_file`, `parent_node=FSl0bCi9raBuLbsdTbHlgb0agwf`, `size`, and the PDF file.58 Capture `data.file_token` from the response. If the upload fails, continue without `file_token`.59605. **BUILD the exact record payload.** Run `scripts/build_resume_screening_payload.js` via61 `code_execute` (language `"javascript"`; NO stdin — inline the data into the code):62 `console.log(JSON.stringify(buildPayload({ mode: "create_record", candidateName, email, jobTitle,63 uploadDateMs, score, strengths, gaps, recommendation, file_token, recordId })))`. Pass `recordId`64 only for an update. Use the returned `lark.path`, `lark.method`, and `lark.body`. If code execution65 is unavailable, follow `references/resume-screening-contract.md` field-for-field.66676. **WRITE the record, then post the summary.** Send the builder's body via `nyxid_proxy` slug68 `api-lark-bot` with its `path`/`method`:69 - create: `POST /open-apis/bitable/v1/apps/<app_token>/tables/<table_id>/records`70 - update: `PUT /open-apis/bitable/v1/apps/<app_token>/tables/<table_id>/records/<recordId>`71 using the resolved app_token/table_id and `body` `{ "fields": {…} }`. THEN post a short candidate72 summary to the chat with `lark_messages_send` (or `reply_with_interaction`). **If no base target73 was resolved OR the write fails, SKIP the Bitable write and STILL post the built summary to the74 chat — never fail the run for a missing base.**75767. **REPORT** one line: candidate name, score, create/update (or "summary-only"), and the returned77 Lark Base `record_id` or the skipped reason.7879## Payload builder reference8081Bundled script: `scripts/build_resume_screening_payload.js`. Modes: `create_record`,82`update_record` (requires `recordId`), and `build_record`/omitted (create by default, update when83`recordId` is present). Fields may be at the root or under `body`; it also accepts `ai`,84`modelOutput`, `screeningResult`, and Groq-style `choices[0].message.content` model responses. The85score gate (`>= 60`) and the `FSl0bCi9raBuLbsdTbHlgb0agwf`/`tblgZgSqmeBag2na` base are baked in.86Output is `skip: true`, `needs_more_information: true` + `missing`, or a record object with87`fields`, `lark.path`, `lark.method`, and the connector-ready `lark.body`. Exact aliases, threshold88behavior, and examples are in `references/resume-screening-contract.md`.8990## Guardrails9192- Only use data the tools actually returned — never invent candidate details, resume text, scores,93 file tokens, record ids, or Bitable responses.94- Prefer running the bundled script; it is the deterministic source of truth for the Bitable payload.95- All credentials and token exchange go through your NyxID-brokered tools; never handle raw secrets.