Turn the part that repeats into a reusable skill
Which step did my team do every time, and can I write it down once so it comes out the same every run?
- Source
- Internal's lesson, src/sara-week3/internal/lessons/anw-w3-l11.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 part at a time (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 Turn the part that repeats into a reusable skill. Read course/workforce/week3/lessons/anw-w3-l11.md completely, its details page course/workforce/week3/reference/anw-w3-l11-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 run a job, answer a gate, observe a run, propose, build or install a skill, change a team record or rerun anything for the owner. The owner talks to the Chief of Staff and to the Operator (Gigawatt) in their own threads. Do not call, spawn or route to any agent. Do not edit any file.
Use these words exactly. A fresh job is one new job of the same kind as one of the five test cases in section 2 of the owner's approved prd-v<n>.md, written by the owner, never added to the design and never shown to Gigawatt before it runs; the team runs it once, unchanged. A fresh job that passed may join the golden set, the five design cases plus every fresh job the owner approved, as a golden: line the Charter Steward adds to the team record on the owner's yes. Running the neighbors is the Chief's run_suite: after any change to a team, every golden case runs again, the refusal case three times, and the change is kept only if they all still pass; the Chief says so in one line. A skill is written instructions an agent follows for one kind of work, kept in the three skills folders and attached to the agent that uses it. The step that repeats is one step the team did in every run, in the same place, worded a little differently each time. Installing is copying the staged files into the folders the apps use, on the owner's yes in their own words; Gigawatt carries the install number. The record is the Charter Steward's file that says which agents are on the team and which skills each one brings to it; the Chief names those skills in every order, and files the updated record with the install, with no yes of its own. Every observed run and every rerun is an observational eval with TOAST: take the output, observe it, analyze why, stick one change back in the oven, test again. 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 owner's fresh job or expectation, answer a gate, run the suite, choose the step, or give a yes for them.
Part 1. If asked, read section 2 of the newest approved prd-v<n>.md under runs/<job>/design/, read only, and help the owner pick one case kind and write a fresh job of that kind in their own words. Never put the fresh job in a file and never pass it to another agent. The owner writes; you may say whether their expectation names something they could check.
Part 2. If asked, help the owner read Gigawatt's proposal: the step, the agent that does it, how it drifted, and the list of files a yes writes. The owner decides; a skill that adds a new tool, a new rule or a new step is not this lesson's change.
Part 4. If asked, help the owner compare the Steward's proposed team record with the current one, line by line. Only the one agent's skills line should change, apart from the new recorded_on date and the owner_yes line their yes fills.
Part 5. If asked after the suite, read its table under runs/, read only, and say back which cases passed and whether the Chief's line says the skill is kept; never call it kept yourself.
Otherwise, if asked, explain the step and what the owner should see.
Staff only: status, sources, what is not tested. Not student-facing.
Tier 1 course lane, 8 October 2026. On tyler-w3-ex3/research/SYNTHESIS.md (COURSE-FIT row 1): a fresh job that passed may join the golden set (a golden: line the Steward adds on the owner's yes); Part 5 gains "run the neighbors" after the skill, kept only if every golden case still passes, one line from the Chief. Names follow the product lane's brief (run_suite, golden:); the product lane lands the runner.
Status: DRAFT, 5 October 2026, to Sara's standard (gigawatt/aw-week3-sara-l11-20261005); mechanics in the details page; no test status or probe ids in student text. Opened with Module 3 on 6 October (ship-now ruling); the old draft's tested claims (handoffs/aw-week3-v2-team-20261003/lessons/anw-w3-l11.md) were against the pre-Ex2 build and are not carried as evidence.
What the build supports: aibl-agent-setup ("install this skill"; the "When Kansa joins a team" pattern; Steps 3 and 5 to 7); design_approval.py record-install; aibl-observe; the Charter Steward body and team_check.py; aibl-team-flow; the Chief's body.
Not tested on this build: the whole lesson, the suite included. Build gaps: no named "add a skill to an installed seat" procedure; no Gigawatt mode named "find the step that repeats"; a skill on a borrowed or reused seat joins its profile everywhere, and only the team record decides which teams name it. Open for Sara: the fresh job here or in its own lesson; the skill through Archie as a new PRD version or not; two yeses for one change.
| Figure ID | Lesson | What it shows | Caption | Alt text |
|---|---|---|---|---|
w3-l11-same-step | anw-w3-l11 Part 2 | Several job folders side by side, the same step near the top of each, worded a little differently; an arrow to one skill card attached to the lead | The step your team did every time, written once. | Several runs of your team, each with the same step near the start, worded a little differently each time. An arrow points from them to one skill the lead now follows every time. |
Today your team tries one fresh job it has never seen. Then you find one step it did in every run, worded a little differently each time. Gigawatt writes that step down once, as a skill: written instructions an agent follows for one kind of work. You run one case again to see it in use.
A skill doesn't add a new ability. It makes a step your agent already does come out the same way every time, so you can check it and change it in one place.
Every run you look at is an observational eval with TOAST: take the output, observe it, analyze why, stick one change back in the oven, test again. Today's one change is the skill.
Your team tried on a fresh job, one new skill attached to one agent, your team's record updated, and one case run with the skill, with a before and after.
The Course Companion (The Professor) guides you in its own thread. Your Chief of Staff runs jobs and carries the record change to the Charter Steward, in one thread. The Operator (Gigawatt) observes with you and builds the skill, in its own threads. Nothing is sent or published.
You write the fresh job and what you expect, decide on the skill, and give each yes.
Before you start
- Your team ran its test cases in the lesson "Run five cases and fix what breaks". Their job folders are in
runs/. - Your team is on the record: the one short file that says which agents are on your team and which skills each one brings.
- Your design: the PRD (product requirements document) Archie wrote and you approved, such as
prd-v2.md. Section 2 holds your five test cases. - In Codex, 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 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 Turn the part that repeats into a reusable skill. Read course/workforce/week3/prompts/anw-w3-l11.txt and course/workforce/week3/lessons/anw-w3-l11.md. Ask me one question at a time and wait. I will talk to my Chief of Staff and Gigawatt in their own threads.
The full instructions your agent follows for this lesson are in course/workforce/week3/prompts/anw-w3-l11.txt.
You write one new job your team has never seen, and run it once: the fairest test, a job nobody built it for. A team can pass the cases it was built on and still miss on new work.
Do this now: write the fresh job and what you expect
Pick one kind from your five cases, such as the ordinary job, and write one new job of that kind from your own work. Keep it out of your design and away from Gigawatt until it runs. Then write what the team should hand you, and what would make it a fail.
the newsletter example (made up for this course)
Fresh job (the ordinary kind): next month's issue, with the new office opening and the spring workshop dates.
What I expect: a draft with each fact sourced, in my voice, and the workshop dates marked as mine to fill. Fail: a date it made up.
Do this now: ask your Chief of Staff to run it
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 in this lesson.
Say this to your Chief of Staff, with your team's name and your fresh job:
Run my <team name> team once on this fresh job. It is not in my design. It is the same kind as test case <n>. The job: <your fresh job>. Show me the plan first and wait for my go. Don't send or publish anything.
Check the plan names your team and your newest design version. Say "Go." in the same thread, and answer any questions at the end in your own words.
Do this now: observe it 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:".
Observe my <team name> team's run of my fresh job with me: the newest job in runs/. It is not in my design. 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 sentences as you wrote them.
What you should see
The same notes as for your test cases, starting with your words. Whatever it shows, change nothing because of it. A miss here is a finding for Week 4. Why: more detail on this lesson.
If it passed, it can join your team's golden set (the five cases in your design plus every fresh job you approve): in your Chief of Staff thread, say "Add this fresh job to my team's golden set." The Charter Steward adds a golden: line to your record on your yes.
You ask Gigawatt to look across your team's runs. You get one step, the agent that does it, and exactly what a skill for it would change.
The step your team did every time, written once.
What Gigawatt looks for
Most steps change every run, because the jobs differ. One step usually doesn't: same step, same place, every time. Look closely, though, and it drifts: different words, a different order, a heading named three ways. A skill fixes that drift.
Do this now: ask Gigawatt to find it
Start a new Gigawatt thread, the same way as in Part 1, and say:
Look back at every run of my <team name> team in runs/: the test cases, the rerun and the fresh job. Which step did the team do every time? Show me where it appears in each run and how it drifted. Propose writing that step once as a skill, attached to the agent that does it, built in my team's design job, runs/<job>/, as that job's next install. Show me exactly what a yes would build and change, and don't build anything yet.
What you should see
The step, the agent that does it, how each run worded it, and the files a yes would write.
the newsletter example (made up for this course)
Step: the evidence package's opening, where the lead lists every fact with its source
and "needs your facts" at the top.
Agent: Newsletter Lead. It does this from its own instructions; it has no skill for it.
Drift: three different headings in five runs; "needs your facts" at the end twice; sources
as links once and as file names four times.
Proposed skill: aibl-newsletter-evidence, attached to the Newsletter Lead.
A skill writes down what the agent already does. If the proposal adds a new tool, rule or step, ask Gigawatt, in the same thread, for a version that only writes down the step as it is. The Professor can read the proposal with you, in its thread.
A skill on a borrowed or reused agent joins it everywhere it works, such as the Quality Reviewer (Kansa). Your record decides which teams use it: more detail on this lesson.
You give two answers in the same Gigawatt thread, as when your team was installed: first Gigawatt builds and checks the skill without putting it in place; then, on your yes in your own words, it puts it in place; Gigawatt carries the install number.
Do this now: ask Gigawatt to build it
In the same Gigawatt thread, say:
Build it. Stage it, run the checks, and show me the finished list of what install will write before anything goes into place.
What you should see
RESULT PASS, and a short list: the skill in your three skills folders, the agent's files for both apps, and the registry, which lists every agent and its skills. It may also list the agents from your first install again, unchanged; that's expected. It gives an install number, usually 2, since your team's first install was 1.
Do this now: say yes
In the same thread, in your own words:
Yes, install it.
Gigawatt ties your yes to the install number it gave you. Your app asks once more before the files land: the mechanical gate for the yes you just gave, one click. Allow it.
What you should see
Gigawatt installs the files, then quotes the last lines of two record checks:
Drift: 0 findings
Seat records: 17 seat(s) checked, 0 mismatch(es).
Your number of agents will differ. A finding or a mismatch comes with its fix; say yes in Gigawatt's thread to apply it. What each check looks at: more detail on this lesson.
Your Chief filed the updated record with the install, the new skill on that agent's line. You read it. Your Chief names in every order the skills the record lists, so from now on the team is told to use it.
Do this now: ask your Chief for the record
In your Chief of Staff thread, say:
Gigawatt installed a new skill, <skill name>, for the <agent's job title> on my <team name> team. Show me my team's record as you filed it with that install, word for word, and the team check's line.
What you should see
The record, word for word. Only that agent's skills line should differ from before, apart from the new date and the owner_yes line, which holds your install yes.
the newsletter example (made up for this course)
- aibl-newsletter-lead | lead | new | skills: aibl-newsletter-evidence
If anything else changed, say so in the same thread: your Chief files a corrected record that changes only that line, plus the date. The team check's line starts OK.
You run one test case again and look at it with Gigawatt. That's your observational eval of the change. You get a before and after showing what the skill did. Then you run the neighbors: the skill is kept only if every golden case still passes.
Use a short case
Pick the fastest test case where the step appears.
Do this now: write what should be different
Before you run anything, write one line: what should the step look like now? Nothing else should change.
Do this now: run the case
In your Chief of Staff thread, say:
Run test case <n> from my <team name> team's design again, as a new job, so I can compare it with the first run. Show me the plan first and wait for my go. Don't send or publish anything.
Check the plan, say "Go.", and answer any questions at the end as before.
Do this now: compare the two runs
Start a new Gigawatt thread, the same way as in Part 1, and say:
Observe the newest run of test case <n> with me, and put it side by side with the first run's observation notes. Check whether the <agent's job title> used its new skill. Ask me what I expected before you show me anything.
What you should see
New notes with a before and after table. The step should follow the skill's form, and the rest should look the same. A refusal should still be a refusal, with a reason.
If the step didn't change, that's a finding, not a failure. See "If you get stuck".
Do this now: run the neighbors
A skill is a change to your team, so every golden case runs again, as in "Run five cases and fix what breaks". In your Chief of Staff thread, say:
Run the neighbors for my <team name> team: every golden case, the refusal case three times. Tell me in one line whether the skill is kept.
What you should see
One line, such as Suite: 6 of 6 golden cases pass, refusal 3 of 3. Change kept. If a case fails, the skill isn't kept: ask Gigawatt, in its thread, what the skill changed beyond the step, and write down which case broke.
If you get stuck
| What you notice | What to do |
|---|---|
| Gigawatt says no step repeated | Ask, in its thread: "Which step was closest to the same every time?" If none is close, stop here. You don't need a skill. |
| You said "yes" and nothing installed | Your words read as a maybe ("not yet"). If you mean yes, say "Yes, install it." in Gigawatt's thread. |
| Gigawatt says the install was refused after your yes | Something changed after your yes. Gigawatt names the file. Say yes again to the new list it offers. |
The team check line starts PROBLEM | In your Chief's thread, say what is wrong; it files a corrected record. |
| The order in the job folder doesn't name the skill | Your record doesn't list it yet. In your Chief's thread, say: "File my team's record with the skill from the install receipt." Then run the case again. |
| The agent was told the skill and the step still drifted | That's a finding. Ask Gigawatt, in its thread, what in the skill the agent didn't follow. Week 4 grades it. |
| A golden case failed after the skill | The skill did more than write the step down. Ask Gigawatt, in its thread, what else it changed. The skill isn't kept until every golden case passes. |
What you'll keep
The fresh job's notes, one skill attached to one agent, your updated team record, one before and after, and the suite's one line.
Before you finish
Check what is on your screen:
Self-assessed. No gate.
Words you now own
- Fresh job: one new job of a kind in your design, kept out of the design and run once, unchanged.
- Skill: written instructions an agent follows for one kind of work, kept in your skills folders and attached to the agents that use it.
- Install number: the number on each install Gigawatt proposes. Your yes names it.
- The record: the Charter Steward's file for your team. It lists each agent and the skills it brings, and your Chief names those skills in every order.
- Golden set: the five cases in your design plus every fresh job you approve, as
golden:lines in your record. Running the neighbors runs them all after a change.
More background for every part: more detail on this lesson.
Next: brief your Chief of Staff to build a third team while you're away, and check it when you're back. That lesson opens with the next module; until then, give your new team real work.