Recruiter Runbook

Ask first

• Which sources may it read? (default your inbox, your texts, and notes from meetings if you keep them; ignore newsletters and automated mail)
• Who counts as someone you owe or who owes you? (default anyone who wrote to you personally in the last 60 days)
• When is a promise stale? (default 30 days, or 60 days if it has no date)
• Where should the list live? (default one page you open each morning)

Needs

• Read access to the sources
• One saved list with these fields for every item: title, who, direction (mine or theirs), source and link, date captured, due date if any, confidence (high or low), status (open, done or dropped) and why it closed
• A scheduler that can run every 15 minutes
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. Every 15 minutes, read messages and notes added since the last run. Pick out sentences where someone commits to do something: "I will", "I'll send", "we'll have it by", "can you send me".
2. Save each as one item with every field above. Mark confidence low if the commitment is vague or the date is a guess.
3. When a later message in the same thread contains the promised thing (an attachment, a link, the words "here it is"), close the item and write why. Never close an item on silence.
4. Mark an item stale after the stale limit and move it to a dropped list, never delete it.
5. Show the list with days waiting, oldest first, amber after seven days. This one list is the source for every other run that needs promises: the morning brief reads its Due from you section from it, and the follow-up run adds each commitment it drafts to it.

Check

• Seed three promises in a test thread; confirm all three appear with the right direction and date
• Send the promised thing; confirm the item closes and says why
• Confirm a vague promise is marked low confidence and a stale one moves to the list of dropped items
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on a 15-minute schedule, one list I open each morning; my morning brief and my follow-up run both read and write it

Ask first

• Where are the compensation bands?
• Who approves an offer? (default you and finance)
• What may it propose? (default only within the band)

Needs

• Read access to the role brief, bands and interview notes
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. When a candidate is marked final round, read the brief, the band and every interviewer's note.
2. List what the candidate said they care about, in their words.
3. Draft an offer inside the band and list questions that are still open.
4. Leave it as a draft with a checklist of approvals. Never send.

Check

• Run it for a past hire; the offer lands inside the band and names what they asked for
• Confirm nothing was sent
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on demand, a draft in my documents

Ask first

• What time should it arrive? (default 7am on weekdays, your time zone)
• Which calendars and inboxes count? (default your work calendar and work inbox)
• Which accounts or customers should it watch? (default any that wrote to you in the last 30 days)
• Where should it land? (default a page you open, or a message from your assistant)

Needs

• Read access to your calendar and inbox
• A short list of accounts to watch, or a way to infer them from recent mail
• The promises list, if you keep one. If you have none, build it by pulling "I will" and "I'll send" lines from your last 14 days of sent mail, or paste one in
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. Each weekday before the set time, read today's events, unanswered mail from the last three days, and any change in the watched accounts since yesterday.
2. Write one page with four parts: Today, Needs you, Moved overnight, Due from you. Read Due from you straight from the promises list; do not rebuild it from mail. Keep each part to at most five lines.
3. If a part is empty, leave it out.
4. Put the page where you read it. Never send anything on your behalf.

Check

• Run it once now and confirm each line names a person or an account, not a general remark
• Confirm nothing on the page is more than three days old unless it is a promise
• Confirm Due from you matches the promises list
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on a weekday schedule at 7am, written to one page I open in the browser

Ask first

• What must a person have? What would you count as a bonus? (default: the role's years of experience, two named skills, and the right to work in the location)
• Where are the applications? (default your hiring tool or the inbox for the role)
• How many on the shortlist? (default five)

Needs

• Read access to the brief and applications, or paste them in
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. Each morning, read the brief and list must-haves and bonuses, then find applications that arrived since yesterday.
2. Read each application and check it against them.
3. Rank with one line of evidence from their own words. Never use age, name or photo.
4. Name one thing to check on the first call. Add anything interviewers flagged in the debrief run for this role, so tomorrow's ranking learns from them.

Check

• Run it on a past role; the shortlist includes the person you hired
• Confirm no reason relies on a protected trait
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on a daily 8am schedule, a page in my notes

Ask first

• Which interviews? (default any on your calendar with a candidate name)
• How long should the pack be? (default one page, under 250 words)
• Who else is interviewing?

Needs

• Read access to the calendar and candidate files
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. At 6pm each evening, find tomorrow's interviews.
2. Read the candidate's file.
3. Write their story, role-based questions and gaps. Include the one thing to check that the shortlist run named for this person.
4. Leave space for notes.

Check

• Run it on a past interview; every fact is in their file
• Confirm no question is about a protected trait
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on an evening schedule, a pack in my notes

Ask first

• Which calls count? (default any call you have notes for)
• Whose voice should the draft use? (default yours, learned from your last 20 sent emails)
• How soon after the call? (default within two hours, checked hourly on weekdays)

Needs

• Your notes from each call, or a written-up recording, and a way to find the email thread for the person
• Permission to create a draft only, never to send
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. Every hour on weekdays, find calendar events that ended in the last two hours and have notes. A call has ended when its calendar event has ended.
2. Find the email thread with the attendees. If there is none, start a new draft addressed to them.
3. From the notes, list what was agreed, who owns each item and by when, and one open question. If the notes say nothing was agreed, say so on the first line of the draft and do not invent next steps.
4. Write a short reply in your voice as a draft on that thread. Never send it.
5. Add each commitment you made in the draft to the promises list, so it is tracked from the moment you hang up.

Check

• Run it on yesterday's call; every commitment in the draft appears in your notes
• Confirm the draft is in drafts and nothing was sent
• Confirm each commitment also appears on the promises list
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on an hourly weekday schedule, a draft left on the existing thread and each commitment added to my promises list

Ask first

• Which meetings count? (default any event with an attendee whose email domain is not yours)
• How long before should it arrive? (default 30 minutes)
• Where should it land? (default a page in your notes, linked from the event)
• Which sources may it read about the attendees? (default your own mail and notes, plus their public profiles)

Needs

• Read access to your calendar, inbox and notes
• A way to write a page and attach its link to the event
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. Every 15 minutes, look at the next 45 minutes of your calendar for events with an outside attendee that have no brief yet. Outside means an email domain that is not yours.
2. For each, find the last three threads and notes involving those people.
3. Write a page: who is in the room, where you left it, what each side owes (read from your promises list if you keep one), anything new about them, one open question.
4. Keep it under 150 words. If you have no history, say so and give one fact about the company.
5. Never contact the attendees.

Check

• Run it on a past meeting you remember; it names the thing you promised and did not do
• Confirm no sentence states something it could not find in your mail or notes
• Confirm an internal-only meeting gets no brief
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on a 15-minute schedule, a page written into my notes thirty minutes before each external meeting

Ask first

• Who is the critic, and in whose voice? (default a demanding magazine editor who speaks to the profile owner directly)
• Who may use it? (default anyone with the link, rate-limited)
• Are profiles written in the third person allowed? (default no; they stop with a zero and a short note)
• What is the scale? (default 1 to 7 with a name per step, most profiles landing at 3 or 4)
• Where do results live? (default a shareable page per result)
• Can you host a small web page and pay for model calls? (default no; then run it in your coding assistant and publish the result as a static page)

Needs

• A web host that runs server code, a key-value store for results, and a model account with a spending cap.
• A rate limiter keyed by visitor address.
• One long persona prompt and a defined output shape.
• If your assistant can only chat: paste the profile text and the persona prompt; it returns the verdict in chat and you share the text by hand.

Build

1. Read: a page where a user pastes profile text. Strip interface noise (buttons, "see more", counts), cap the length, rate-limit by visitor address.
2. Decide (persona): the critic judges how engaging the profile is, not how impressive the person is, speaking to the owner as "you". Define the scale in the prompt: 1 Threadbare, 2 Tepid, 3 Pedestrian, 4 Flickering, 5 Noteworthy, 6 Formidable, 7 Unforgettable, with what a 7 requires. Fame earns nothing.
3. Decide (output): structured data: full name; a third-person flag with a short monologue; rating and its name; a one-sentence verdict; story, personality and memorable, each scored 1 to 5 in half steps with a headline and a short body; a few numbered suggestions. An About section in the third person returns a zero. The three sub-scores must agree with the overall rating; if they do not, the model is asked once more.
4. Deliver: stream the answer so the rating and verdict appear first. Recover from malformed output by trying stricter parsing in turn; if all fail, show a retry message. Save each result under a name-based address so it can be shared. Fallback: the verdict as plain text.
5. Memory: the saved results and an error log in the store. One model call per profile; cap monthly spend.
6. Optional: a route where the coding assistant takes the persona prompt itself and publishes the page, so the model bill is avoided.

Check

• Paste a made-up profile: a rating, verdict and three sub-scores in the right voice, and the page opens at its own address after a refresh.
• Paste one written in the third person: flagged with a zero.
• Quiet run: no traffic means nothing is written; the first real use shows as a saved result.

In my version

• Profiler, a site I built, with the warmth of Miranda Priestly. Serverless host, key-value store, streamed answers, one address per result, and a terminal route that runs on my subscription instead of the model bill.

Ask first

• Where do interviewers write notes? (default a shared document or the hiring tool)
• What decision does the debrief feed? (default hire, no hire, second round)
• How long should it be? (default one page, under 250 words)
• When does it run? (default 30 minutes after the last interview note is added)

Needs

• Read access to the notes, or paste them in
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. After a round, collect every interviewer's notes.
2. List points where they agree and points where they differ, quoting each.
3. Recommend a next step and say what it rests on. Pass what interviewers flagged to the shortlist run for this role.

Check

• Run it on a past round; the disagreements are real
• Confirm no recommendation rests on a protected trait
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code after each round, a page in my notes

Ask first

• Who is the list?
• What do you want to say about the role? (default two lines)
• Who approves? (default you)

Needs

• Read access to the brief and list, and permission to create drafts, never to send
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. For each person, find one real thing from their public work.
2. Write three short sentences: what I noticed, the role, the ask.
3. Leave drafts. Never send.
4. Skip anyone who has opted out.

Check

• Draft three for a test list; each mentions something real
• Confirm nothing was sent
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on demand, drafts in my mail

Ask first

• Where are your past applicants and notes?
• How far back? (default 18 months)
• Who must approve before anyone is contacted? (default you)

Needs

• Read access to past applicants and notes, or paste a list
• If your assistant cannot read one of these, paste it when it asks. A thinner version still works
• Any password or token is entered by you into your device's secure store, never pasted into a chat or saved in a plain file

Build

1. When a role is opened, read the brief and list its must-haves.
2. Search past applicants for people who met most of them but were passed on for a different role or reason.
3. For each, say why they were passed on, and why they might fit now.
4. Never contact anyone. Never use age, name or photo.

Check

• Run it for a role you filled; a person you wish you had remembered appears
• Confirm no reason relies on a protected trait
• Say whether the writing step ran for real or on sample text. Check that every figure, quote and link matches its source, and drop any that does not

In my version

• Claude Code on demand, a short list in my notes