# Interview Followup

> Use right after an interview, when a user says "I just finished the interview", "help me write a thank-you note", "debrief me", or "the [company] round is done". Drafts a 24-hour thank-you note (writing-quality gated), runs a structured debrief, and closes the learning loop by feeding new questions/answers back into the story bank, application record, and tracker. After the interview, and the no-recording path, working from memory. Before it, use `interview-prep`. With a transcript, use `interview-debrief`.

- Skill: `guerrilla2799/interview-followup` (Agent Skill)
- Install (CLI): `npx skillmds@latest add guerrilla2799/interview-followup`
- Raw SKILL.md: https://api.skillmd.com/api/skills/guerrilla2799/interview-followup/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: guerrilla2799 (https://skillmd.com/u/guerrilla2799)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/guerrilla2799/interview-followup

---


# Interview Followup

The 24-hour follow-up plus the debrief learning loop. Two jobs: send a sharp thank-you fast, and make sure this interview makes the next one better. Most candidates send the thank-you and skip the debrief. The debrief is where each interview improves the next.

**Boundary with `interview-debrief`:** use *this* skill when there's **no** recording. It works from the user's memory of the round. If the user recorded the interview and has a transcript, use `interview-debrief` instead. It does a deeper, evidence-based analysis and drafts the follow-ups from what was actually said.

## When to use
- "I just finished my interview", "help me write the thank-you", "debrief me on that round"
- Ideally within a few hours of the interview, while it's fresh. The thank-you is time-sensitive (aim for <24h) and the debrief details fade fast

## Inputs
- Reads: `pipeline/applications/<id>-...md` (the role, the prep pack, interviewer names), `workspace/output/<id>-interview-prep.md` (what you expected to be asked), `workspace/stories/story-bank.md`, `templates/outreach-templates.md` (#5, the post-interview thank-you)
- Needs from user: who they met (names/titles if not already recorded), and their memory of the conversation, which the debrief questions below pull out

## Workflow

### Part 1: Thank-you note (fast, gated)
1. **Gather the specifics.** Ask for the interviewer name(s) and one genuinely interesting thing from the conversation. The note is worthless if it's generic. It must reference something real that was actually discussed.
2. **Draft from template #5** in `templates/outreach-templates.md`: thanks, one specific point from the conversation, one line reinforcing fit, and, if a question was fumbled, one clean sentence that answers it better (see Part 3). Under ~120 words. One note per interviewer if there were several, and personalize each with something distinct to that person, never a copy-paste.
3. **Run the `writing-quality` skill as a gate.** This is mandatory, because the thank-you is buyer-facing writing. Strip AI-slop tells ("I hope this finds you well", "I wanted to reach out"), keep it human and short. Do not present the draft until it has passed this gate.
4. **Save the draft** to `output/<id>-thankyou-<interviewer>.md` and present it for the user's approval. **Never auto-send.** The user sends it themselves after reviewing.

### Part 2: Structured debrief (the learning loop)
5. Run the debrief while it's fresh. Ask these, conversationally, and capture the answers:
   - **What questions were you actually asked?** (Especially any you didn't expect. Those are gold for the bank.)
   - **What went well?** Which answers landed, what did they react positively to?
   - **What did you fumble?** Where did you ramble, blank, or give a weak answer?
   - **What did you learn about the role/team/company?** New facts about the job, the manager, the priorities, the process, comp, timeline.
   - **What's your gut read?** How did the room feel: engaged, skeptical, going through the motions?
   - **Likelihood?** Their honest sense of advancing, and what the interviewer said about next steps/timing.

### Part 3: Close the loop (this is the point)
6. **Feed the story bank.** For every new or unexpected question, append it to `stories/story-bank.md`. If it exposed a competency the bank doesn't cover, flag it as a gap. If the user fumbled a question, draft the **stronger answer now** (STAR, mapped to a real story) and store it with the question so next time it's ready. This is how `story-bank` compounds across a search.
7. **Draft stronger answers for fumbles.** For each fumbled question, write the crisp answer the user wishes they'd given. Optionally fold a one-sentence version into the thank-you note (Part 1). It is a graceful way to correct a weak moment while it still matters.
8. **Update the application record** `pipeline/applications/<id>-...md`: append to the Debrief section (date, round, questions asked, what went well, what to fix, gut read, next steps the interviewer named) and note the thank-you was drafted.
9. **Update the tracker** `pipeline/tracker.csv`: advance `status` (e.g. `interviewing` → `final`, or hold at `interviewing` if more rounds), set `next_action` and `next_action_date` (send the thank-you today, and record the expected decision or next-round date), and update `notes` with the one-line read.
10. **Report** the loop: thank-you path (awaiting the user's send), what got appended to the bank, and the tracker change.

## Output
- Writes to:
  - `workspace/output/<id>-thankyou-<interviewer>.md` – the drafted note(s), `writing-quality`-passed, **not sent**
  - `workspace/stories/story-bank.md` – appended new questions + stronger answers (+ any new gap flags)
  - `workspace/pipeline/applications/<id>-...md` – Debrief section updated
  - `workspace/pipeline/tracker.csv` – status, next_action, next_action_date, notes updated
- Format: the thank-you follows template #5, the debrief follows the application-record Debrief section, and bank additions follow `templates/story-template.md` (or a compact Q→better-answer pair where a full STAR entry isn't warranted)

## Rules & quality bar
- **Thank-you within 24h, and specific.** A generic note is worse than none. It must reference something that was genuinely discussed.
- **`writing-quality` gate is mandatory before any note is shown.** No exceptions. This is a writing-producing skill.
- **Never auto-send.** The human reviews and sends every note themselves. Drafts are saved and presented. Sending is the user's action.
- **No fabrication.** Don't invent enthusiasm, praise, or a conversation detail that didn't happen. If the user can't recall a specific, keep the note shorter rather than padding it.
- **The debrief must close the loop, not just record it.** The debrief matters because new questions land in the story bank and fumbles get a stronger answer before the next round, so the next interview starts ahead. Skipping Part 3 defeats the skill.
- **Be honest in the debrief.** If the gut read is bad, say so and reflect it in the tracker. A realistic pipeline beats an optimistic one.

## Related skills
- Reads from: `interview-prep` (what was expected vs. what happened), `story-bank`, `company-research`
- Runs: `writing-quality` (mandatory gate on the thank-you note)
- Feeds: `story-bank` (new questions plus better answers, the compounding asset), `application-tracker` (status + next action), and improves the next `interview-prep` for later rounds
- Escalates to: `offer-negotiation` when a debrief signals an offer is coming