follow-up

name	follow-up
description	Surfaces follow-up priorities across all active applications using company-specific response window data. Classifies each application as within window, approaching, overdue, or critical. Researches and stores response window data for any company not yet on file. Builds FollowUpBrief entries and invokes writer for send-ready messages. Connector context enriches briefs with email/calendar data when available and approved.
triggers	["follow up on my applications","what needs a follow-up","should I follow up","what should I follow up on","follow-up queue","any follow-ups due","check my follow-ups","should I follow up on","how long has it been since I applied","am I being ghosted","any overdue applications","what's taking so long"]

Surface follow-up priorities for all active applications using company-specific response window data.

Data files

File	Purpose
{user_dir}/CareerNavigator/tracker.json	Active applications with stage history and dates
{user_dir}/CareerNavigator/company-windows.json	Per-user company response window data, researched on demand
{user_dir}/CareerNavigator/profile.md	Target companies, industries, and role context

---

Workflow

1. Load active applications

Read tracker.json. Collect all applications where status is not accepted, rejected, withdrew, or ghosted. These are the applications that need follow-up tracking.

For each active application, extract: company, role, date_applied, status, stage_history (most recent entry date), follow_up_date, contacts.

2. Load or initialize company-windows.json

Read {user_dir}/CareerNavigator/company-windows.json. If it does not exist, initialize it:

{
  "meta": {
    "created": "{today}",
    "version": "1.0",
    "description": "Company-specific hiring response windows for {user's name} — researched from public sources and stored per session"
  },
  "companies": {},
  "size_tier_fallbacks": {
    "startup": {
      "typical_first_response_days": { "min": 3, "max": 14 },
      "follow_up_after_days": 10,
      "notes": "Startups move fast; silence beyond 2 weeks usually means no."
    },
    "mid_market": {
      "typical_first_response_days": { "min": 7, "max": 21 },
      "follow_up_after_days": 14,
      "notes": "Mid-market timelines vary; a single polite follow-up at 2 weeks is standard."
    },
    "enterprise": {
      "typical_first_response_days": { "min": 14, "max": 42 },
      "follow_up_after_days": 21,
      "notes": "Large companies have multi-layer review queues; 3–6 weeks before initial screen is normal."
    }
  }
}

3. Research missing companies

For each active application's company that is not already present in company-windows.json.companies, research it now.

Research method:

Use web search to find candidate-reported hiring timelines for this specific company and role level. Search for:

• "{company name}" hiring process response time site:glassdoor.com
• "{company name}" recruiter response time PM OR director site:linkedin.com
• "{company name}" interview process how long Blind OR Reddit

From the results, extract:

• Typical days from application to first recruiter contact (range)
• The point at which following up is appropriate (not too early to seem anxious, not so late it's futile)
• Any ATS or process notes specific to this company (e.g., uses Greenhouse, has structured panels, known for slow enterprise hiring)

Store the result in company-windows.json:

"Anthropic": {
  "researched_at": "{today}",
  "typical_first_response_days": { "min": 14, "max": 35 },
  "follow_up_after_days": 21,
  "size_tier": "startup",
  "process_notes": "Uses Greenhouse. High application volume for PM/research roles. Recruiter screen typically 3–5 weeks. Structured multi-round process. Following up via Greenhouse portal or LinkedIn is appropriate at 3 weeks.",
  "sources": ["Glassdoor reviews", "LinkedIn recruiter posts", "Blind threads"]
}

If web search returns insufficient data for a specific company, fall back to the matching size_tier_fallbacks entry and note it:

"Acme Corp": {
  "researched_at": "{today}",
  "typical_first_response_days": null,
  "follow_up_after_days": 14,
  "size_tier": "mid_market",
  "process_notes": "No company-specific data found — using mid-market defaults.",
  "sources": ["size_tier_fallback"]
}

Write the updated company-windows.json before proceeding.

4. Classify each application

For each active application, calculate:

• days_elapsed = today − date_applied
• last_activity = date of the most recent stage_history entry
• days_since_activity = today − last_activity
• window_max = company-windows.json → typical_first_response_days.max for this company

Classify:

Status	Condition
within_window	days_elapsed ≤ window_max
approaching	days_elapsed > 75% of window_max and ≤ window_max
overdue	days_elapsed > window_max and ≤ 1.5× window_max
critical	days_elapsed > 1.5× window_max

Applications at phone_screen or interview stage use days_since_activity for the overdue check rather than days_elapsed from application — these have a shorter expected cadence (typically 3–7 days between touches).

5. Check interview-specific follow-ups

For any application with a stage_history entry where stage contains "interview" and the entry date is within the last 7 days, check whether a thank-you note has been logged (look for a notes entry with "thank-you" or "thank you" or "follow-up" after the interview date). If not:

• If interview was 0–2 days ago: flag as thank_you_due
• If interview was 3–7 days ago: flag as thank_you_overdue

6. Check offer deadlines

For any application where offer.deadline is set, calculate days remaining. Flag as offer_deadline_approaching if ≤ 3 days remain.

7. Output the follow-up report

Order: critical → offer deadlines → thank-you due/overdue → overdue → approaching → within window.

**Follow-up Queue** — {today's date}

{critical count} critical  ·  {overdue count} overdue  ·  {approaching count} approaching

---

🔴 CRITICAL  ({days_elapsed}d elapsed · window was {window_max}d)
  {Company} — {Role}
  Applied {date_applied}. Last activity: {last_activity_description}.
  {1 sentence on what action makes sense now — follow up, move on, or both}

🟡 OVERDUE  ({days_elapsed}d · window {window_min}–{window_max}d)
  {Company} — {Role}
  {1 sentence context}

📬 THANK-YOU DUE
  {Company} — {Role}  (interview {n} days ago)

🟢 APPROACHING  ({days_elapsed}d · window closes in ~{n}d)
  {Company} — {Role}
  {follow_up_after_days} day follow-up point: {follow_up_date}

✓ WITHIN WINDOW
  {Company} — {Role}  (day {days_elapsed} of {window_max})

If the queue is empty (all active applications are within window and no thank-yous outstanding):

"All active applications are within their expected response windows. Nothing to follow up on yet."

8. Follow-up copy via writer

For each application classified as overdue or critical (and thank_you_due / thank_you_overdue), do not draft send-ready text in this skill.

8a. Build FollowUpBrief entries

For each row that needs a message, add an object to a list:

• company, role
• stage_kind: post_application | thank_you_post_interview | offer_nudge | other (infer from stage_history)
• channel: email default; linkedin if user or notes indicate
• recipient_name, recipient_title from contacts[] if present—else null and use neutral greeting instruction
• applied_or_event_date, days_elapsed (or interview date for thank-yous)
• facts_hooks: 1–3 bullets from notes / stage_history.post_notes—no invention
• tone: direct, not sycophantic; 3–5 sentences target for email body
• connector_context: optional short inbox/calendar context when connectors are available and the user approved lookup; else null

Read {user_dir}/CareerNavigator/profile.md and voice-profile.md for sign-off name and voice.

Voice preflight (once per session before 8b): If you will invoke writer and voice-profile.md lacks substantive pasted prose under ## User writing samples or ## User writing samples (launch), ask the user for 2–5 LinkedIn posts or short writing (or skip for neutral tone). Append pastes to voice-profile.md. If samples already exist or the user skips, proceed.

8b. Invoke writer

• Use the exact agent name writer with mode follow-up. Pass all FollowUpBrief objects in one invocation when possible. Retry once on failure.

8c. Present output

Under each application entry, show writer output (subject line if email, body, LinkedIn variant if requested).

When email/calendar connectors exist, enrich briefs with approved prior-thread context before invoking writer.

8d. Sent confirmation + auto-track

After presenting all messages, say:

"Reply 'sent [Company]' for each message you send and I'll log it to your tracker."

When the user confirms a send for a specific company:

• Find the matching application record in tracker.json
• Append to contacts[].interactions[] for the relevant contact (or the first listed contact if ambiguous):

  { "date": "YYYY-MM-DD", "type": "email | linkedin", "notes": "Follow-up sent — {stage_kind}" }
  

• Append to notes[]: { "date": "YYYY-MM-DD", "text": "Follow-up sent via {channel}: {stage_kind}" }
• Update follow_up_date to today + the company's follow_up_after_days window (use size-tier fallback if not found)
• Update next_step to "Await response to follow-up"
• Confirm: Logged: follow-up sent to {Company} — next follow-up window: {new follow_up_date}

---

What You Never Do

• Do not mark an application as overdue before its research-backed window has elapsed — do not use the flat 7-day rule
• Do not draft send-ready follow-up prose in this skill—only FollowUpBrief objects; writer writes the message
• Do not draft a follow-up for an application still within its window
• Do not fabricate contact names — use "there" if no contact is on file
• Do not store company window data outside of {user_dir}/CareerNavigator/company-windows.json
• Do not re-research a company already present in company-windows.json unless the entry is more than 90 days old