Candidate Intake

Candidate Intake

Turn applicants who arrive through many different channels into one running, scored master Google Sheet, then help the user pick who to contact and send the outreach. This is a recruiting funnel in a box: it does not matter whether someone applied on Indeed, emailed a resume, commented on a Facebook job post, or sent a Messenger DM. They all land in the same sheet, scored the same way, so the user has a single ranked view of every candidate.

  Indeed -+
  Gmail  -+
  FB post-+--> normalize --> score --> dedupe -->  ONE master Google Sheet
  FB DM  -+                                          |
  paste  -+                                          v
                                    review ranked list --> approved outreach

The guiding idea: organize and score everyone first, then let the user decide who gets outreach. Never message people before the user has seen the ranked list and chosen. That review gate is the whole point, so respect it.

Prerequisites

This skill works almost entirely through the browser, so a community member can use it with zero API keys or developer setup.

• Claude in Chrome extension connected. Gmail, Facebook, Indeed, and Google Sheets are all read and written through Chrome. If it is not connected, ask the user to connect it before proceeding.
• A Google account with the applicant channels logged in in that Chrome profile (Gmail, and Facebook if they use it).
• A master Google Sheet. Setup helps the user create one, or point at an existing sheet.
• A config file, .candidate-intake-config.json, saved in the working folder. It holds the role definition, scoring rubric, sheet URL, and outreach templates so every run and the daily sweep behave consistently.

Step 0: Setup (run once, or when the role changes)

Before the first intake, check the working folder for .candidate-intake-config.json.

• If it is missing, run setup: read references/setup-and-config.md and follow it. Setup collects the role and company, the must-have and nice-to-have criteria, the screener questions and their ideal answers, the scoring weights, the sender's first name, and the outreach templates. It then creates (or adopts) the master Google Sheet and writes the header row. Save the result to .candidate-intake-config.json using assets/config.template.json as the shape.
• If it exists, load it and continue. If the user says the role or criteria changed, re-run the relevant part of setup and save.

Do not hardcode a specific role into the workflow. The rubric is configurable on purpose so the same skill works for a cold caller, a VA, an acquisitions rep, a dispositions manager, or anything else the community hires for.

The intake loop

Run these in order. Use a task list so the user can follow along.

1. Load config. Confirm the role and the master sheet URL out loud in one line so the user knows what they are adding to.
2. Pick channels. Ask which sources to pull this run, or default to every channel the config has enabled. Channels are independent; do whichever the user has new applicants in.
3. Gather raw applicants from each chosen channel. Each channel has its own reference with exact steps:
   • Indeed -> references/intake-indeed.md
   • Gmail -> references/intake-gmail.md
   • Facebook group post + Messenger -> references/intake-facebook.md
   • Pasted email or text -> references/intake-manual.md
4. Normalize every applicant into the Candidate Record (schema below). Collect them into a single JSON array for the run. Never invent data: if a field is not present in the source, leave it blank.
5. Score the array with scripts/score_candidates.py using the rubric in references/scoring-rubric.md. This adds a numeric score, a tier (Interview / Maybe / Dismiss), and a short reason to each record. Scoring is deterministic so the same candidate always lands the same way.
6. Dedupe and append. Read references/sheet-schema.md. Export the current sheet rows, run scripts/prepare_rows.py to drop anyone already in the sheet (matched on email, phone, or profile handle) and build the paste block, then write the new rows into the master sheet through Chrome. Verify the row count went up by what you added.
7. Present the ranked review. Show the user a tight, ranked summary of who is new this run, grouped by tier, with the one-line reason and the channel each came from. Then ask, in one question, who they want to reach out to. Do not send anything yet.
8. Send approved outreach. For the candidates the user picked, send the screening message on the channel that candidate came in through, following references/outreach.md. Verify each message before sending.
9. Offer the daily sweep. Once the user has done this a couple of times, offer to schedule a daily run that checks Gmail and Facebook for new applicants, appends them, and reports a short digest with the top new qualifiers flagged for approval. See references/setup-and-config.md for the scheduling snippet.

Candidate Record schema

Normalize every applicant, from every channel, into these fields. Consistency is what lets one sheet hold everyone.

Hard rules

These are non-negotiable because they came directly from how the user works.

• Never use em dashes in any outreach. Not one, ever. Write with commas, periods, or short sentences instead. This is a strict brand rule. Proofread each message for the em dash and en dash characters before sending.
• Keep outreach casual and first-name only. Open with "Hey {first_name},". No formal salutations, no long paragraphs.
• Verify before every send. Read the compose box back (screenshot or page text) and confirm the name matches the conversation before hitting send. Messaging the wrong person is the easiest mistake to make and the hardest to take back.
• Never fabricate. Do not guess emails, phone numbers, resume details, or screener answers. A blank field is fine and honest; an invented one is not.
• Dedupe before you append. The same person often applies twice, sometimes on two different channels. Match on email, phone, or profile handle and update the existing row rather than creating a duplicate.
• Respect the review gate. Score and log everyone first. The user chooses who to contact. Outreach is never automatic unless the user has explicitly turned that on for the daily sweep.

Reference map

Read the file for the step you are on. Do not load them all up front.

Scripts

• scripts/score_candidates.py reads a JSON array of Candidate Records plus the config, writes back the same array with score, tier, and score_reason filled in. Deterministic weighted scoring.
• scripts/prepare_rows.py takes the scored records and a CSV export of the current sheet, drops anyone already present, and prints a tab-separated block ready to paste into Google Sheets, plus a summary of what was added vs skipped.

Both are plain Python 3 with no dependencies beyond the standard library, so they run anywhere.