Developer Handover Pack

Developer Handover Pack

Job Posting & Applicant Tracking

Listens for a Greenhouse webhook firing when a vacancy is marked as approved. Pulls the vacancy record fields (role title, department, hiring manager, salary band, and stored job description template) from Greenhouse via the Jobs API, passes them to a prompt template stored in the orchestration layer, and generates a complete, on-brand job advertisement. The draft is dispatched to the hiring manager's Gmail inbox as a structured email containing the ad copy and a single approval link. On approval the agent triggers the job board posting workflow. On rejection it loops back with the hiring manager's revision notes attached.

// Input
greenhouse.job.id          : string   // Greenhouse internal job ID
greenhouse.job.title       : string   // Role title
greenhouse.job.department  : string   // Department name
greenhouse.job.description : string   // Raw JD template stored in Greenhouse
greenhouse.job.hiring_mgr  : string   // Hiring manager email address
greenhouse.job.salary_band : string   // Salary range if public-facing

// Output (on draft sent)
gmail.message.to           : string   // hiring manager email
gmail.message.subject      : string   // '[Action Required] Job Ad Draft: {job.title}'
gmail.message.body         : string   // Generated ad copy + approval/reject links
gmail.message.sent_at      : datetime // ISO 8601 timestamp

// On approval
greenhouse.job.ad_approved : boolean  // Flag set to true on hiring manager click
workflow.next_trigger      : string   // 'post_to_linkedin' event emitted

Monitors Greenhouse for newly submitted application records on any active role. On each new application event it reads the full applicant payload (name, email, phone, resume URL, source channel, and applied-at timestamp), writes a structured record to Greenhouse via the Applications API if not already complete, appends a corresponding row to the live Google Sheets hiring tracker, and dispatches a personalised acknowledgement email via Gmail within five minutes of the application being received. The agent includes a dedupe check against the Greenhouse candidate ID to prevent duplicate records on re-submissions or webhook retries.

// Input
greenhouse.application.id          : string   // Greenhouse application ID (used for dedupe)
greenhouse.application.candidate_id: string   // Greenhouse candidate ID
greenhouse.application.job_id      : string   // Linked job ID
greenhouse.application.first_name  : string
greenhouse.application.last_name   : string
greenhouse.application.email       : string
greenhouse.application.phone       : string
greenhouse.application.resume_url  : string   // Hosted resume link from Greenhouse
greenhouse.application.source      : string   // e.g. 'LinkedIn', 'Direct'
greenhouse.application.applied_at  : datetime // ISO 8601

// Output (Greenhouse write-back)
greenhouse.application.status      : string   // Set to 'active'
greenhouse.application.tag         : string   // 'intake_complete'

// Output (Google Sheets append)
sheets.row.job_title               : string
sheets.row.candidate_name          : string
sheets.row.email                   : string
sheets.row.phone                   : string
sheets.row.resume_url              : string
sheets.row.source                  : string
sheets.row.applied_at              : datetime
sheets.row.status                  : string   // Default: 'Applied'

// Output (Gmail acknowledgement)
gmail.message.to                   : string   // applicant email
gmail.message.subject              : string   // 'We received your application for {job_title}'
gmail.message.body                 : string   // Personalised template with estimated response timeline
gmail.message.sent_at              : datetime

Triggers when an application record in Greenhouse is tagged 'intake_complete' by the Application Intake Agent, confirming the record is fully populated. Reads the resume content from the Greenhouse-hosted URL and evaluates it against the role's defined screening criteria stored as a custom field set on the Greenhouse job record. Assigns a numeric fit score (0 to 100) and a short reasoning note (maximum 80 words). Writes the score, reasoning, and a candidate tag (shortlist, hold, or reject) back to the Greenhouse application record. After all applications for a role batch are scored, compiles a ranked shortlist and sends a Slack notification to the hiring manager's Slack user ID with candidate names, scores, and direct Greenhouse profile links.

// Input
greenhouse.application.id             : string   // Application to score
greenhouse.application.resume_url     : string   // Hosted resume document
greenhouse.job.screening_criteria     : string   // Custom field: structured criteria stored on job record
greenhouse.job.id                     : string
greenhouse.job.title                  : string
greenhouse.job.hiring_mgr_slack_id    : string   // Slack user ID for notification

// Output (Greenhouse write-back per application)
greenhouse.application.fit_score      : integer  // 0 to 100
greenhouse.application.score_note     : string   // Max 80 words reasoning
greenhouse.application.tag            : string   // 'shortlist' | 'hold' | 'reject'
greenhouse.application.scored_at      : datetime // ISO 8601

// Output (Slack shortlist notification, sent once per role batch)
slack.message.channel                 : string   // Hiring manager Slack user ID (DM)
slack.message.text                    : string   // Ranked candidate list with scores and links
slack.message.blocks                  : array    // Structured Block Kit payload
slack.message.sent_at                 : datetime

// ============================================================
// TRIGGER: Greenhouse vacancy approved
// ============================================================
EVENT: greenhouse.webhook -> job.approved
  payload.job_id          : string   // e.g. 'gh_job_00412'
  payload.job_title       : string   // e.g. 'Senior Marketing Manager'
  payload.department      : string   // e.g. 'Marketing'
  payload.hiring_mgr_email: string   // e.g. 'jordan@yourcompany.com'
  payload.hiring_mgr_slack: string   // e.g. 'U04XKJW29AB'
  payload.salary_band     : string   // e.g. '$70,000 to $85,000'
  payload.jd_template     : string   // Raw JD from Greenhouse custom field

// ============================================================
// AGENT 1: Job Ad Drafting Agent
// ============================================================
AGENT_1_INPUT:
  job_id, job_title, department, hiring_mgr_email, salary_band, jd_template

STEP 1.1: LLM prompt with stored ad template + job fields
  -> ad_draft.body        : string   // Generated ad copy
  -> ad_draft.subject     : string   // Email subject line

STEP 1.2: Gmail send to hiring_mgr_email
  gmail.to                = hiring_mgr_email
  gmail.subject           = '[Action Required] Job Ad Draft: {job_title}'
  gmail.body              = ad_draft.body + approval_link + reject_link
  approval_link.token     = HMAC(job_id + timestamp, secret_key) // TTL 72 hrs

STEP 1.3: Await webhook callback from approval link
  IF approved:
    greenhouse.job.ad_approved = true
    EMIT event: 'post_to_linkedin'
  IF rejected:
    hiring_mgr.revision_notes -> append to prompt -> regenerate (max 2 loops)
  IF timeout (72 hrs):
    SEND Gmail reminder -> IF no response after 24 hrs: ALERT Slack @hiring_mgr

STEP 1.4: LinkedIn post (on approval)
  linkedin.job_post.title     = job_title
  linkedin.job_post.body      = ad_draft.body
  linkedin.job_post.location  = greenhouse.job.location
  linkedin.response.post_url  -> written to greenhouse.job.board_links

// ============================================================
// AGENT 2: Application Intake Agent
// ============================================================
TRIGGER: greenhouse.webhook -> application.created
  payload.application_id  : string
  payload.candidate_id    : string
  payload.job_id          : string
  payload.first_name      : string
  payload.last_name       : string
  payload.email           : string
  payload.phone           : string
  payload.resume_url      : string
  payload.source          : object   // -> extract source.public_name
  payload.applied_at      : datetime

STEP 2.1: Dedupe check
  QUERY greenhouse.applications WHERE candidate_id = payload.candidate_id
    AND job_id = payload.job_id
  IF record exists:
    LOG 'duplicate_skipped' -> error_log table
    HALT

STEP 2.2: Greenhouse write-back (if new)
  greenhouse.application.status = 'active'
  greenhouse.application.tag    = 'intake_complete'

STEP 2.3: Google Sheets append (Sheets API v4)
  sheets.spreadsheet_id   = config.SHEETS_HIRING_TRACKER_ID
  sheets.range            = 'Applications!A:K'
  sheets.row              = [
    job_title, candidate_id, first_name + ' ' + last_name,
    email, phone, resume_url, source.public_name,
    applied_at, 'Applied', '', ''
  ]

STEP 2.4: Gmail acknowledgement send
  gmail.to      = payload.email
  gmail.subject = 'We received your application for {job_title}'
  gmail.body    = template.acknowledgement(first_name, job_title)
  gmail.sent_at -> logged to error_log on failure

// ============================================================
// AGENT 3: Resume Screening Agent
// ============================================================
TRIGGER: greenhouse.application.tag == 'intake_complete'
  application_id, resume_url, job_id, job_title, hiring_mgr_slack_id

STEP 3.1: Fetch screening criteria
  QUERY greenhouse.jobs[job_id].custom_fields.screening_criteria
  -> criteria_array : array  // JSON array of weighted criteria strings

STEP 3.2: Fetch and parse resume
  HTTP GET resume_url -> raw document (PDF | DOCX)
  PARSE -> resume_text : string

STEP 3.3: LLM scoring call
  INPUT: resume_text + criteria_array
  OUTPUT:
    fit_score  : integer  // 0 to 100
    score_note : string   // max 80 words
    tag        : string   // derived from thresholds in config store
      70-100 -> 'shortlist'
      40-69  -> 'hold'
      0-39   -> 'reject'

STEP 3.4: Greenhouse write-back
  greenhouse.application.fit_score = fit_score
  greenhouse.application.score_note = score_note
  greenhouse.application.tag        = tag
  greenhouse.application.scored_at  = NOW()

STEP 3.5: Batch check
  QUERY greenhouse.applications WHERE job_id = job_id
    AND tag NOT IN ['shortlist','hold','reject']
  IF count == 0 OR elapsed_since_first_score >= 4 hrs:
    COMPILE ranked_shortlist (tag='shortlist', sorted by fit_score DESC)
    SEND slack.chat.postMessage to hiring_mgr_slack_id
      blocks: [candidate_name, fit_score, tag_emoji, greenhouse_profile_url]
  ELSE:
    WAIT (next application.scored event)