Brief your Chief of Staff to build a third team while you're away, and check it when you're back
Can my organization build a team while I'm away, and how do I check it when I'm back?
- Source
- Internal's lesson, src/sara-week3/internal/lessons/anw-w3-l10.md (Internal b518c35b, 2026-10-22 17:15), read by md_week3.py.
- Copy
- Current Markdown is vendored byte for byte from the Internal commit in Source by vendor_week3.py. Historical publication records: 10-06 Exercise 1 with Sara's yes; 10-07 Exercise 2 and the Module 2 send-route and follow-up fixes, student 58c60d50; 10-08 Exercise 3 and the l05/l11 diagram fixes, student 55061386. Those records do not qualify later Outlook changes. Module 3 and 4 drawings are rendered from the Mermaid in the lesson file (md_week3.py --figures).
- Prompts
- Word for word from the lesson file Internal ships (course/workforce/week3/). Not run end to end from Learn.
- Checked in the app
- Not checked in the app from Learn.
- Open to students
- Yes.
- Other view
- one scroll (one part at a time is the lesson; one scroll is kept to compare)
Staff review only: full controlling prompt. Students paste only the short request below.
Use the native aibl-the-professor agent to guide the Week 3 lesson Brief your Chief of Staff to build a third team while you're away, and check it when you're back. Read course/workforce/week3/lessons/anw-w3-l10.md completely, its details page course/workforce/week3/reference/anw-w3-l10-details.md when asked for more, and the native aibl-course-guide and aibl-lesson-map methods for this app. You teach and read; you never brief the Chief, write the overnight contract, run the plan, the dry run or the start, observe a case, approve a design, install, record or stop anything for the student. The student talks to the Chief of Staff and the Operator (Gigawatt) in their own threads. Do not call, spawn or route to any agent. Do not edit any file. Never write, suggest or pass on a yes; every yes is the student's own words in the right thread.
Use these words exactly. A background build (the course first called it an overnight build) is the organization designing, staging, checking and rehearsing a new team in the background while the student is away; the student starts it at any time of day and checks it when they are back, and overnight is only one example. Nothing is installed, recorded, sent or published while they are away. The contract is the six files the Chief saves before the start under runs/<job>/overnight/contract/: the brief, the visible test cases, the sealed case, the sources the agents may read during the run, the tools and the limits. The plan id is the twelve characters the plan prints; the Chief records it with the student's yes, and the student never types it. The dry run is one short session started the way the real run will start; it must pass before the start. The sealed case is one of test cases 2 to 4 that the Chief keeps apart, and nobody sees it until the last step of the run. Staged means prepared and checked, not installed. The status line is the first two lines of the status the student reads when they are back: the state, and whether the protected files are unchanged. The report is runs/<job>/overnight/morning-report.md; below its fixed line are notes a model wrote. Observing a case is an observational eval with TOAST: take the output, observe it, analyze why, stick one change back in the oven, test again. The student says yes at two doors: the design, before they go, to the brief, the cases and the plan the Chief shows them, in their own words; and the install, when they are back, in Gigawatt's thread, in their own words, Gigawatt carrying the design version and the install number. The pre-authorization is one line they may add at the first door, "If every case passes, install it; only stop me for a Needs-you item"; then the install goes ahead without a second yes when every check and every case, the sealed one included, passed, and waits otherwise. The team record and the daily check are filed with the install, with no yes of their own. Between the doors the agents stop only for the smoke alarm: money, sending outside, deleting, posting publicly, account authority; a stop is a park on the Needs-you list with the work saved, never a halt. Name agents job title first, nickname in brackets on first use. Say "your work", never "business"; refer to other lessons by title, never by number.
Ask open questions only. Never write the student's brief, test cases or expectations, never say the pre-authorization line for them, and never choose whether to install. Never ask them to type a plan id, a version or an install number.
Part 1. If asked, help the student put the work, who receives it, how often and their starting point in their own words, and read course/workforce/week3/team-shape.md with them. They write; you may say whether a test case has Input, Should and Fails if lines they could check.
Part 2. If asked, read the plan and the dry run's last line as the student shows them, and say back plainly what runs while they are away and what never does, and what the pre-authorization line would and would not do. Never write their yes.
Part 3. If asked, read runs/<job>/overnight/morning-report.md, read only, and help the student find the status line, the Needs you list, the seats marked ready for your yes, each case's row, anything refused during the run, the decisions and assumptions listed for them, and the cost line (This run used about N tokens (about $X); if more than they expected, lower limits in the next plan). Say which part a model wrote. Help them write one line per case of what they expect before Gigawatt shows them anything.
Part 4. If asked, read the card and the assumptions in the newest runs/<job>/design/prd-v<n>.md, read only, with the student before they say yes to the install in Gigawatt's thread, or after a pre-authorized install, so they know what was installed.
Otherwise, if asked, explain the step and what the student should see.
Staff only: status, sources, what the build supports. Not student-facing.
Status: SHIP, 6 October 2026, to Sara's standard, on the overnight build v2 (handoffs/aw-week3-ex3-20261005/); Tyler and Sara ruled to ship it now and tweak it after it is live. Each part opens with what the student does and gets; mechanics live in course/workforce/week3/reference/anw-w3-l10-details.md; no test status, probe ids or lesson numbers in student text or the paste prompts.
What the build supports: aibl-overnight-team (contract, plan, dryrun, the yes naming the plan id, start, status, report, stop, --resume); workforce/house/overnight.py v2; the house settings (start is an ask rule); Archie's and Gigawatt's overnight paragraphs; design_approval.py; the Steward's team record from runs/<job>/setup/team-record.proposed.md. The detail is in handoffs/aw-week3-ex3-20261005/.
Framing (Tyler's ruling, 7 October): background autonomy, not only overnight: the team builds while the student is away, started at any time; "overnight" stays only as an example and in file and command names. Windows: start refuses this week; the Chief offers the daytime build instead.
Convergence pass, 8 October (lane C, student text only): the three readiness rules in Before you start with the Chief's refusal; the worst-case length on the plan; the dry run's check of an agent already on the record; the report leads with Needs you, then the agents ready for your yes, then the concerns that ride along; a first build may end with one or two Needs you items. Tier 1 lane, 8 October: the cost line in Part 3 and the details page.
| Figure ID | Lesson | What it shows | Caption | Alt text |
|---|---|---|---|---|
w3-l10-overnight | anw-w3-l10 Part 2 | A timeline from "before you go" to "when you're back", with a locked box labeled "never while you're away: install, record, send"; the alt text lists each step | Your organization works while you're away. Everything that needs your yes waits for you. | A timeline. Before you go, you give your Chief of Staff the brief, read the plan, and say yes naming its plan id. While you're away, Archie designs the team, Gigawatt prepares each new agent, an independent check reviews each one, and your test cases run on the prepared agents. Installing, recording and sending never happen while you're away. When you're back, you read the status, observe the results with Gigawatt, and say yes to the design, the install and the record. |
You built your second team step by step, with Archie and Gigawatt. This time you describe a third team to your Chief of Staff, answer its questions, read the plan, say yes, and step away. Start it whenever suits you. It runs in the background.
While you're away, your organization designs the team, prepares the new agents, checks each one, and runs your test cases on them. One case stays sealed until the very end, so the team gets one fair test it was never built on. When you're back, you read what happened, look at the results with Gigawatt, and decide.
Nothing is installed, recorded or sent while you're away. Your yes before you go is the first door: the design. The install is the second, when you're back: one yes in your own words, or none, if you said before you left that the team may be installed when every case passes. The record and the daily check are filed with the install. Between your two yeses the team stops only for five things: money, sending anything outside, deleting, posting publicly, and account authority. Each one waits on your Needs-you list with the work saved up to that point; everything else keeps going.
A third team, designed and tested without you watching, your notes on each test case, and the team installed and on the record.
An organization that only works while you watch saves you very little. This is the first time yours works on its own, inside limits you set and a plan you approved.
The Course Companion (The Professor) guides you in its own thread. First you talk to your Chief of Staff. When you're back, your Chief of Staff briefs you, and the Operator (Gigawatt) looks at the results with you and installs the team. The Charter Steward's record is filed by your Chief with the install.
You describe the work, write the test cases, and give your two yeses in your own words: the design before you go, the install when you're back.
Before you start
- Your second team is on the record. You finished the lesson "Turn the part that repeats into a reusable skill".
- Pick the work. One recurring piece of your own work that fits the five roles in
course/workforce/week3/team-shape.md, like your second team. - Your material is filed and approved. The files the team should read are filed as source cards by the Knowledge Librarian (Sallee), as in "Process the material through Cairns", and approved for the team's department. If they aren't yet, your Chief tells you which files are missing, and Sallee files and approves them as part of your yes to the plan.
- Hatch's source map names your work. The Internal Knowledge Researcher (Hatch) reads only the places on it, set up in "Set up your email department". If it names the course's example, ask your Chief for the tools interview again, or say the team reads files only.
- Your org chart before the new team is saved. You ask for it at the start of Part 1; the Charter Steward saves it under
runs/. - Your Chief checks those three before it plans. If one is missing, it tells you the one thing to fix, and carries the missing files into your one yes.
- Your starting point: how long the work takes you now, and how often it slips.
- While it runs, your computer stays on. Plugged in, awake, and the app open, for a few hours. On a laptop, leave the lid open.
- In Codex, you never talk to an agent directly: Codex passes your words on. Start your first message to an agent in a thread with "Ask the <agent> to do this, and show me its reply word for word:". Start every message after that, your answers included, with "Tell the <agent> this, word for word, and show me its reply word for word:".
Start or resume this lesson
Open your prepared workbench and call The Professor by its role name, aibl-the-professor. Paste only this.
Use the native aibl-the-professor agent to start the Week 3 lesson Brief your Chief of Staff to build a third team while you're away, and check it when you're back. Read course/workforce/week3/prompts/anw-w3-l10.txt and course/workforce/week3/lessons/anw-w3-l10.md. Ask me one question at a time and wait. I will talk to my Chief of Staff, Archie and Gigawatt in their own threads.
The full instructions your agent follows for this lesson are in course/workforce/week3/prompts/anw-w3-l10.txt.
You tell your Chief of Staff what the new team should produce, and answer its questions in one message. You get your org chart before the new team, a saved brief, and your five test cases.
You don't design the team yourself this time; you say what you want it to produce. Your Chief checks the three things above and tells you what is missing, then asks at most five questions, in one message. Answer what you can, and say "use your best judgment, marked as an assumption" for the rest. When you're back you see every assumption.
Do this now: print your org chart first
In Claude Code: start a new thread, type @ and pick aibl-chief-of-staff. In Codex: start a new thread and begin with "Ask the Chief of Staff to do this, and show me its reply word for word:". Use this thread for every Chief step before the start.
Say this to your Chief of Staff:
Have the Charter Steward print my registry, so I have my org chart before the new team.
Do this now: brief your Chief of Staff
In the same thread, say this, with your own work:
Build a third team in the background while I'm away. Read course/workforce/week3/team-shape.md first. The work: <what it makes, who receives it, how often>. Ask me everything you need in one message.
Answer its questions in one message, in the same thread: your starting point, what the specialist does that none of your agents can, where the material is and which tools the work lives in, your five test cases, and anything else it asks. Your Chief asks once whether it may look through those tools to see how you do this work now; say yes if you're happy for it to. It isn't a new door.
the partner outreach example (made up for this course)
Starting point: about 2 hours a month, and I skipped 3 of the last 6 months.
Specialist: judges whether a partner fits my customers, my $150 budget and my rules. None of my agents does that.
Material: my partnership notes and my list of local businesses, both in my library.
Tools: my Gmail. Yes, you can look through it.
Cases: 1, pick November's partner and draft the pitch. 2, the same, for owners of senior dogs.
3, a pitch to a new cafe I know nothing about. 4, the festival date, where my notes and the town calendar disagree.
5, a partnership to promote calming chews. The team must refuse.
Use your best judgment on anything else, marked as an assumption.
What you should see
The Steward's printout of your registry. Then the readiness check: READY, or what is missing, named in one line each (a file Sallee will file and approve on your yes, or the one thing only you can fix). Then a short list of questions, and after your answers, the brief your Chief saved, with the five roles (which ones are new) and the assumptions it will work from, and your test cases written out with Input, Should and Fails if lines. Your Chief keeps one of cases 2 to 4 sealed: nobody building the team sees it, and it runs only at the very end. Its facts may still be in your material; that's fine.
You read the plan your Chief shows you and watch a short dry run. You say yes, in your own words: the first door. You get the build running in the background; nothing is installed, recorded or sent until the second door, when you're back.
Your organization does the work in the boxes while you're away.The diamonds wait for you. Installing, recording and sending never happen while you're away.
Four steps while you're away, one yes when you're back
The plan lists what runs, in order, the longest it can take at worst, and what never happens while you're away. The new agents are staged: prepared and checked, and loaded only into the build's test run, not installed.
The dry run is a short practice start, under a minute in Claude Code and a minute or two in Codex. It starts one session the way the real run will, checks that everything the run needs is reachable and that a write outside the job's folder is refused, and grades one agent you already have, to show the check passes a good agent. Your Chief starts the run only after it passes.
Do this now: read the plan
Your Chief shows you the plan right after it saves your brief, then runs the dry run. If it stops before the plan, say, in the same thread:
Show me the plan, then run the dry run.
Read the plan: check that the four steps, the worst-case length and the "never happens" line are there. The line Plan id: is your Chief's to carry: your yes is recorded against it, and you never type it.
Do this now: say yes, in your own words
When the brief, the cases and the plan look right and the dry run says PASS, say this in the same thread:
Looks good, go.
"Yes" works the same way. This one yes covers the design, the build, the checks, the test run, and the filing of any files your Chief named as missing. Your app asks once before the run starts: the mechanical gate for the yes you just gave, one click. Allow it. In Codex, it asks to run the command outside its sandbox, for the dry run and for the start: approve both.
Do this now: decide about the install, now or when you're back
Your Chief offers one line, once. It lets Gigawatt install the team when you're back without asking you, as long as every test case passed, the sealed one included, and nothing needs you. If you want that, say it in your own words; if you'd rather say yes yourself when you're back, say nothing, and it stays off.
If every case passes, install it; only stop me for a Needs-you item.
What you should see
Dry run: PASS, the plan can be started.
Started the background build for <job> (process <number>). It runs while you're away. Keep the app open and the computer awake, plugged in and open.
Leave the app open and the computer on. Come back in a few hours.
You ask your Chief how the build went, then look at each test case with Gigawatt. You get the status line, the report, and notes on each case starting with what you expected. That's an observational eval with TOAST: take the output, observe it, analyze why, stick one change back in the oven, test again.
Do this now: ask for your brief
Start a new Chief of Staff thread, the same way as before, and say:
I'm back. How did the build go? Brief me.
What you should see
The status line comes first:
Background build <job>: done; ...
Protected files: unchanged at each check (<number> files)
"Protected files: unchanged" means nothing an install or a record would change was touched while you were away. Then the report, morning-report.md. It leads with Needs you: anything only you can change, like a source, a rule, a borrowed agent's limits or a case, and any smoke-alarm item parked with its saved work. Then the new agents marked ready for your yes, each with any concerns that ride along. Then the cases, each with a receipt; whether the sealed case stayed sealed; and the decisions and assumptions waiting for you. Then the limits, with one line to read: This run used about N tokens (about $X). If that's more than you expected, lower the limits in the plan next time. Last come the run notes, below a fixed line. A model wrote those, so read them as notes, not as the record. If you pre-authorized the install and every case passed, the report also says the team was installed, with the three checks that let it through.
Expect one or two Needs you items on a first build. Nobody could ask you for a name, an amount, or which of two dates is right. That's the team being honest, not broken.
Do this now: write what you expect
Before you look at any result, write one line per case: what should the team hand you, and what would make it a fail? Your test cases' Should: and Fails if: lines are a good start. Put them in your own words.
Do this now: observe the cases with Gigawatt
In Claude Code: start a new thread, type @ and pick aibl-gigawatt. In Codex: start a new thread and begin with "Ask the Operator (Gigawatt) to do this, and show me its reply word for word:". Say:
Observe the test cases my background build ran in runs/<job>/rehearsal/, one at a time. Ask me what I expected before you show me anything. Then write the observation notes and the list of what isn't mine yet.
Give it your lines as you wrote them.
What you should see
Notes for each case that start with your words, then what each agent produced, then what isn't yours yet. The case your team must refuse should be refused, with a reason. More on reading the report and the notes: more detail on this lesson.
You read the design and say yes to the install, in Gigawatt's thread: one yes, as for your second team. You get the team installed, checked and on the record, with your daily check turned on, and your org chart after it.
If you pre-authorized the install before you left, and the report says the team was installed, skip to "What you should see" below: Gigawatt did this part on your line, and the report quotes the three checks that let it through. If the report says the install waited, it names which check failed; read on.
Do this now: read the design and say yes, in Gigawatt's thread
Start a new Gigawatt thread, the same way as in Part 3, and say:
Install what my background build staged in runs/<job>/setup/. First show me the design it was built from: the card, every assumption marked for me, and the test cases. Then run the checks again and show me what the install writes.
Read each assumption and the install list. If they're right, say this in the same thread:
Yes, install it.
That one yes covers the design and the install: Gigawatt records which version you approved and which install it was, and you never type either. Your app asks once more before the files land: the mechanical gate for the yes you just gave. Allow it.
If an assumption is wrong, tell Gigawatt what, in the same thread. A wording fix is drift: Gigawatt notes it and goes on. A change to what the team may do (a tool, a write path, sending or posting, a borrowed agent, a case) goes to Archie as a new version, and Gigawatt rebuilds what changed with you, as in the lesson "Build the team with Gigawatt", then asks for your yes again.
What you should see
Gigawatt installs the files, then quotes the last lines of two record checks:
Drift: 0 findings
Seat records: <number> seat(s) checked, 0 mismatch(es).
Then the team's record, filed by your Chief with this install, your yes written in, and the team check's line:
OK work/course/staff/teams/<id>.md: 5 members, five roles filled, design found
This is your third team, so the same filing turns on the Charter Steward's daily check: from now on, before your first brief each day, the Steward checks your organization and proposes anything that needs fixing. Gigawatt then asks you to prove each new agent in a new thread, as for your second team.
Do this now: print your org chart after the new team
In your Chief of Staff thread from Part 3, say:
Show me my third team's record word for word, then have the Charter Steward print my registry again, so I have my org chart after the new team.
What you should see
The record, with the five roles, each agent's kind (new, reused or borrowed), and the design line naming the version Gigawatt installed from. Then your registry printout, with your new agents. If a line of the record is wrong, say what, in the same thread: your Chief files a corrected record.
If you get stuck
| What you notice | What to do |
|---|---|
The dry run says FAIL | Don't start. Ask your Chief, in its thread, what the dry run found and what to change. A changed brief or case means a new plan and a new plan id. |
| Your Chief asks for your yes again | Your words read as a maybe ("not yet", "hold on"), or the plan changed and your Chief shows a new one. If you mean yes, say "Yes." |
| In Codex, it says "Not started: running inside Codex's sandbox" | The approval to run outside the sandbox didn't happen. Ask your Chief to start it again, and approve it. |
| You need to stop it | Say "Stop the background build." in your Chief's thread. |
The status says incomplete | A step ended early, or the computer slept. Ask your Chief what finished and what's under Needs you, then to carry on. It needs your new yes. |
The status says blocked | A decision only you can make stops everything, or a fix didn't take. Read BLOCKED.md with your Chief, decide, then carry on with a new yes. |
The status line says VIOLATION | Something changed a protected file while you were away. Don't install. Ask your Chief which files changed, and decide what to do. |
An agent is marked needs you | Don't install yet. Fix what its Needs you line names, then ask your Chief to carry on. If the agent itself failed, fix it with Gigawatt, in its thread, in the daytime. A pre-authorized install never goes ahead with an agent in this state. |
| A case needs a change | Tell Gigawatt in its thread; it goes to Archie as a new version. The install waits for your yes to it. |
| Gigawatt installed without asking you | You pre-authorized it before you left, and every check and case passed; the report quotes them. If you didn't, stop and tell Gigawatt in its own thread. |
| You're on Windows | The background build isn't available on Windows yet. Your Chief offers to run the same build with you in the daytime, as for your second team. |
What you'll keep
Your third team, installed and on the record, the build's brief and report, your notes on each test case, and your org chart before and after. Week 4 starts from these.
Before you finish
Check what is on your screen:
Self-assessed. No gate.
Words you now own
- Background build: your organization designing, preparing, checking and testing a team in the background while you're away, day or night, and stopping before anything that needs your yes. Overnight is one good time, not the only one.
- Plan id: the twelve characters your Chief's plan shows. Your Chief records your yes against it, so it covers exactly that plan.
- Two doors: the two moments you say yes: the design before you go, the install when you're back.
- Pre-authorization: the one line you may add at the first door so the install goes ahead when every case passes. Off unless you say it.
- Dry run: a short practice start that must pass before the real one.
- Sealed case: one test case kept apart from everyone building the team, run only at the end.
- Staged: prepared and checked, but not installed.
- Status line: the first lines of the status you read when you're back: how the run ended, and whether your protected files are unchanged.
- Needs you: a finding only you can change, listed first in the report.
More background for every part: more detail on this lesson.
Next: get ready for Week 4.