Skills workshop · Step-by-step exercise guide

Before you start · choose one app

Choose Claude or Codex with the buttons above. You only need to complete one route in class. The other is available to repeat at home.
Download exec-summary.zip, star-feedback.zip, test-inputs.md and scorecard.md. The ZIP links download files, not websites.
Keep this guide open in a separate browser tab. Open your chosen app beside it, or switch between windows as you work.
Prepare a place to save three answers: Without skill, With skill and Edited skill. A text note is enough. The checkboxes and scorecard on this page save progress in this browser when storage is available; they do not save the AI’s answers.
Run the baseline before installation. Stay with the same app and model. A Claude-versus-Codex comparison changes the model as well as the context.

Checkpoint You have all four files, this guide, your chosen app, and somewhere to keep your outputs. Allow extra setup time on a new account; do setup before class if possible.

Follow the Claude route.Browser · upload ZIP · edit locally · re-upload

1. Open Claude and check access

Open a browser and go to claude.ai. Sign in.
Open your profile or account menu, then Settings.
Open Capabilities. Turn on Code execution and file creation. If it is already on, leave it on.
Open Customize → Skills. Free accounts are supported; an organization can still restrict access.
If exec-summary or star-feedback is already listed, turn its toggle off for now. Leave unrelated settings the same throughout the comparison.
Return to the chat screen. Choose an ordinary chat outside an existing Claude Project, so project instructions do not affect this exercise.

Checkpoint You can open the Skills page. Neither workshop skill is enabled. If access is blocked, use the troubleshooting section or pair with a neighbour.

2. Run the task without the skill

Click New chat. Do not use an old conversation.
Look at the model selector and note the selected model in your notes. Keep that same model for the next run.
Use Copy Pedala prompt below. Click the empty message box and paste with ⌘V on Mac or Ctrl+V on Windows.
Check that the message starts “Summarise these notes for my manager:” and includes all the call notes. Do not add formatting instructions or a skill name.
Click the send arrow. Wait until the answer is finished.
Copy the answer into a note called Without skill, or keep this conversation open. Do not rely on remembering what it said.
Use the scorecard below. Read the answer against each check; record the five scores in the Without column for Claude.

Read the exact pedala prompt

Summarise these notes for my manager:

Call notes– Pedala (e-bike subscriptions, Barcelona) – Tuesday
- attendees: Marta (CEO), Jordi (ops), us
- ~1,200 active subscribers; monthly churn up from 4% to 7% since June!!
- Jordi thinks it's repair wait times: avg 6 days now vs 2 days in spring. Their mechanic left in May
- Marta more worried about new competitor BiciYa, launched at €29/month vs Pedala €39/month
- exit survey: 140 responses, 58% of churned users mention repairs, 22% mention price
- runway ~9 months
- options discussed: hire 2 mechanics (~€6k/month total), cut price to €32, or partner with a repair shop in Gràcia (no quote yet)
- Marta wants a decision before the board meeting on 15 Nov
- we owe them a recommendation by next Friday
- someone mentioned Valencia expansion but parked

(Pedala and BiciYa are fictional companies made up for this exercise.)

Checkpoint You have one saved baseline answer and a score out of 5. You have not uploaded or enabled exec-summary yet.

3. Upload and enable exec-summary

Return to Customize → Skills.
If an exec-summary entry already exists, do not upload a duplicate. Ask a helper to preserve and remove that workshop copy before continuing. If it is a personal skill you need to keep, pair with someone else for this route.
Click +, then Create skill, then Upload a skill. Labels can vary slightly between builds.
In the file picker, open Downloads or the folder where you saved the kit.
Select exec-summary.zip, then click Open to upload. Choose the ZIP, not the extracted Markdown file.
Wait for upload or any organization scan to finish. If an error appears, read it before retrying.
Find exec-summary in the skills list. Switch its toggle on. Keep star-feedback off.

Checkpoint exec-summary appears in the Skills list and is enabled. A successful upload alone is not the same as an enabled skill.

4. Repeat the exact task with the skill

Click New chat again. Do not continue the baseline chat.
Check that the model matches your baseline. Keep other settings unchanged.
Copy and paste the same Pedala prompt using the button below. Do not add “use the skill” on this first attempt.
Send it and wait for the answer to finish.
If Claude exposes its work trace, expand it and look for a read of exec-summary/SKILL.md. Some builds do not expose every file read; record unknown if you cannot tell.
Save the answer as With skill. Fill the With column in the Claude scorecard.
Compare the two answers: recommendation, three reasons, word count, source numbers and final next step. A tie is a valid result.

Read the exact the same pedala prompt

Summarise these notes for my manager:

Call notes– Pedala (e-bike subscriptions, Barcelona) – Tuesday
- attendees: Marta (CEO), Jordi (ops), us
- ~1,200 active subscribers; monthly churn up from 4% to 7% since June!!
- Jordi thinks it's repair wait times: avg 6 days now vs 2 days in spring. Their mechanic left in May
- Marta more worried about new competitor BiciYa, launched at €29/month vs Pedala €39/month
- exit survey: 140 responses, 58% of churned users mention repairs, 22% mention price
- runway ~9 months
- options discussed: hire 2 mechanics (~€6k/month total), cut price to €32, or partner with a repair shop in Gràcia (no quote yet)
- Marta wants a decision before the board meeting on 15 Nov
- we owe them a recommendation by next Friday
- someone mentioned Valencia expansion but parked

(Pedala and BiciYa are fictional companies made up for this exercise.)

Checkpoint You have two answers from fresh chats, the same model and prompt, and two scores. Record whether you observed a skill-file read separately from the score.

5. Open the skill file on your laptop

Open Finder → Downloads on Mac or File Explorer → Downloads on Windows. Find exec-summary.zip.
Mac: double-click the ZIP. Windows: right-click it, choose Extract All, then Extract.
Open the extracted folders until you see the file SKILL.md. The folder immediately containing that file should be called exec-summary.
Mac: right-click SKILL.md → Open With → TextEdit (or a plain-text editor). Windows: right-click → Open with → Notepad. Do not use Word.
You should see the raw text, including ---, name: and description: near the top. Keep those header lines unchanged.

Checkpoint You are editing the actual SKILL.md file in the extracted exec-summary folder, not the ZIP or a formatted preview.

6. Change 120 to 60, then save

Find the exact line - 120 words maximum in total. Count before you answer. under Rules.
Replace only 120 with 60. Leave the rest of the file unchanged.
Save with ⌘S on Mac or Ctrl+S on Windows. If TextEdit offers a rich-text format, choose Format → Make Plain Text before saving.
Confirm the filename is still SKILL.md, not SKILL.md.txt or SKILL.md.rtf. On Windows, enable View → Show → File name extensions if the ending is hidden.
Close the editor, then reopen the file. Check that the saved line now says 60 words maximum.

Checkpoint The saved source file contains the 60-word rule. Changing a local file has not yet updated the skill uploaded to Claude.

7. Upload the edited version

In Finder or File Explorer, go up one level so you can see the exec-summary folder. Select the folder itself, not only SKILL.md.
Mac: right-click → Compress “exec-summary”. Windows: right-click → Compress to ZIP file, or Send to → Compressed (zipped) folder.
Rename the new archive exec-summary-60.zip so you can distinguish it from the original download. Leave the folder inside it named exec-summary.
Inspect the new ZIP. It should contain exec-summary/SKILL.md. Open the file inside it if needed to confirm the 60-word rule is present.
Back in Claude, open Customize → Skills → exec-summary. Disable it, then use … → Delete and confirm removal of that workshop skill. Your local source file and ZIP remain on your laptop.
Use + → Create skill → Upload a skill again. Select exec-summary-60.zip. Wait for completion and enable the uploaded skill.

Checkpoint Claude has one enabled exec-summary entry containing your new version, rather than two competing copies.

8. Test your one-line edit

Start another New chat. Use the same model and the same Pedala prompt.
Send it and save the answer as Edited skill — 60 words.
Count all words in the answer, including the labels. You can paste the answer into a word-count tool, or use the count-only prompt below as a quick check and verify its result.
Compare the answer with the earlier 120-word version. Did it follow the shorter limit and still preserve the key recommendation?
Record what changed in the Exercise 2 note. Do not reuse the original word-limit score as the only check; this time your target is 60.

Read the exact pedala prompt

Summarise these notes for my manager:

Call notes– Pedala (e-bike subscriptions, Barcelona) – Tuesday
- attendees: Marta (CEO), Jordi (ops), us
- ~1,200 active subscribers; monthly churn up from 4% to 7% since June!!
- Jordi thinks it's repair wait times: avg 6 days now vs 2 days in spring. Their mechanic left in May
- Marta more worried about new competitor BiciYa, launched at €29/month vs Pedala €39/month
- exit survey: 140 responses, 58% of churned users mention repairs, 22% mention price
- runway ~9 months
- options discussed: hire 2 mechanics (~€6k/month total), cut price to €32, or partner with a repair shop in Gràcia (no quote yet)
- Marta wants a decision before the board meeting on 15 Nov
- we owe them a recommendation by next Friday
- someone mentioned Valencia expansion but parked

(Pedala and BiciYa are fictional companies made up for this exercise.)

Read the exact word-count check

Count the words in the assistant's summary immediately above, including its labels. Report the count only. Do not rewrite the summary.

Checkpoint You changed the stored instructions, reinstalled them, and checked the new behavior in a fresh chat.

9. Test when the skill activates

Keep the edited exec-summary enabled. Open New chat, paste the Peru prompt below, and send.
Inspect the visible work trace if available. Record Yes, No or Unknown for whether the skill was read. The prediction is No.
Open another New chat. Paste the article prompt below, including its full article, and send. Prediction: Yes.
Open a third New chat. Paste the Barcelona poem prompt and send. Prediction: No.
Record all three observations. Do not put the tests in one conversation, and do not force a skill name into these test prompts.

Read the exact peru test

What's the capital of Peru?

Read the exact article test

Can you condense this article into something my boss can read in 30 seconds?

Campus bike repair pilot

The student services team tested a bike repair desk for 4 weeks. It served 80 students, and 60 booked a second visit. Average turnaround was 2 days, compared with 5 days at the previous provider. The pilot cost €1,600. The team has a €2,000 budget for next month. It has not yet checked whether the provider can handle twice as many bookings. Ana will collect a capacity quote by Friday. The committee must decide on Monday whether to extend the pilot for another month.

Read the exact poem test

Write a poem about Barcelona.

Checkpoint You tested three separate requests and distinguished expected activation from what you could actually observe.

Follow the Codex route.Local folder · install extracted skill · review the file edit

1. Open a separate local practice folder

Open the desktop app you use for Codex and sign in. In builds that label the desktop app ChatGPT, choose its Codex view. This route uses a local folder, not a cloud task or an ordinary ChatGPT conversation.
On your Desktop or in Documents, create an empty folder called Skills Workshop. Do not use a work repository or the club website folder.
Copy the four kit downloads into that folder: exec-summary.zip, star-feedback.zip, test-inputs.md and scorecard.md. Keep both ZIPs uninstalled for the baseline.
In Codex, use its open-folder or project picker to select Skills Workshop. The exact picker label varies by app build. Confirm the folder shown for your chat is this local folder.
Start a New chat in that folder. Paste the availability-check prompt below, then send it. This checks for a pre-existing workshop skill without opening its full instructions.
If exec-summary or star-feedback is already installed globally or through a plugin, stop and ask a helper to disable that copy before the baseline. A new empty folder does not hide user-wide skills. Restart after changing skill settings.

Read the exact availability check

For this practice folder, list the names and paths of any available skills named exec-summary or star-feedback. Do not read their full instructions or summarize any workshop files. If either is already installed, tell me its location before we run the baseline. Do not change anything.

Checkpoint The chat uses your separate Skills Workshop folder, and no active workshop skill is available yet. If the app cannot open local folders, use the optional CLI route at the end or pair with someone whose app can.

2. Run the baseline before installing

Start a New chat in the same practice folder. Keep the model and reasoning setting visible in the composer unchanged for all comparison runs; note them down.
Copy the Pedala prompt below. Paste it into the empty composer with ⌘V on Mac or Ctrl+V on Windows.
Check that it contains the whole call note and no $exec-summary mention. Send the message.
Wait for the final answer. Save it in a note called Codex — Without skill.
Complete the Without column of the Codex scorecard below. Do not compare its score with Claude’s score as a test of the skill; compare within Codex.

Read the exact pedala prompt

Summarise these notes for my manager:

Call notes– Pedala (e-bike subscriptions, Barcelona) – Tuesday
- attendees: Marta (CEO), Jordi (ops), us
- ~1,200 active subscribers; monthly churn up from 4% to 7% since June!!
- Jordi thinks it's repair wait times: avg 6 days now vs 2 days in spring. Their mechanic left in May
- Marta more worried about new competitor BiciYa, launched at €29/month vs Pedala €39/month
- exit survey: 140 responses, 58% of churned users mention repairs, 22% mention price
- runway ~9 months
- options discussed: hire 2 mechanics (~€6k/month total), cut price to €32, or partner with a repair shop in Gràcia (no quote yet)
- Marta wants a decision before the board meeting on 15 Nov
- we owe them a recommendation by next Friday
- someone mentioned Valencia expansion but parked

(Pedala and BiciYa are fictional companies made up for this exercise.)

Checkpoint Your baseline answer and score exist before the skill is installed.

3. Ask Codex to install the local skill

Start a separate New chat in Skills Workshop for setup. Copy the install prompt below and send it.
Read any requested action before approving it. The intended changes are a local Git initialization if needed and the practice folder’s .agents/skills files. No remote, account connection or global installation is required.
Wait for the folder tree. Check for exactly .agents/skills/exec-summary/SKILL.md, with no extra nested exec-summary folder. A dot at the start of .agents makes it hidden in many file browsers.
Open the shown SKILL.md file or its diff to confirm it contains the original 120-word limit. Do not change it yet.
Start a fresh Codex chat in the same folder. If skill discovery has not updated, fully quit and reopen the app, then reopen Skills Workshop and start a new chat.

Read the exact local-install prompt

In this practice folder only, install the workshop executive-summary skill.
1. Confirm this is the Skills Workshop folder I selected. If it is not, stop and tell me.
2. If it is not already a Git repository, initialize it with git init. Do not create a commit or remote.
3. Extract exec-summary.zip. Copy the included exec-summary folder into .agents/skills/ so the final file is .agents/skills/exec-summary/SKILL.md. Do not create a second nested exec-summary folder.
4. If that destination already exists, stop and tell me rather than overwriting it.
5. Show the final folder tree. Do not write a summary, change other settings, or install the skill globally.

Checkpoint The skill exists in the project’s discovery folder. The setup conversation is not used for the measured summary run.

4. Repeat the identical summary task

In the fresh chat, select the same model and reasoning setting as your baseline.
Paste the exact same Pedala prompt below, without adding a skill mention. Send it.
Look in the visible activity for a read of .agents/skills/exec-summary/SKILL.md. Record activation as unknown if the UI does not expose enough evidence.
Save the final answer as Codex — With skill. Complete the With scorecard column.
If it did not activate, keep that result. In a separate fresh chat, try an explicit request beginning $exec-summary followed by the same Pedala prompt. In the Codex CLI or IDE, /skills or typing $ also exposes the skill selector. This forced run is troubleshooting, not the identical-prompt comparison.

Read the exact the same pedala prompt

Summarise these notes for my manager:

Call notes– Pedala (e-bike subscriptions, Barcelona) – Tuesday
- attendees: Marta (CEO), Jordi (ops), us
- ~1,200 active subscribers; monthly churn up from 4% to 7% since June!!
- Jordi thinks it's repair wait times: avg 6 days now vs 2 days in spring. Their mechanic left in May
- Marta more worried about new competitor BiciYa, launched at €29/month vs Pedala €39/month
- exit survey: 140 responses, 58% of churned users mention repairs, 22% mention price
- runway ~9 months
- options discussed: hire 2 mechanics (~€6k/month total), cut price to €32, or partner with a repair shop in Gràcia (no quote yet)
- Marta wants a decision before the board meeting on 15 Nov
- we owe them a recommendation by next Friday
- someone mentioned Valencia expansion but parked

(Pedala and BiciYa are fictional companies made up for this exercise.)

Checkpoint You have the two main answers and scores. Any explicitly forced run is saved separately.

5. Change the instruction in the installed file

Open a separate New chat in Skills Workshop for editing.
Copy the edit prompt below and send it. Codex will edit the installed local file; no ZIP upload is needed.
Open the changed file or the diff shown by Codex. Confirm the only instruction change is 120 → 60 in the word-limit line. The name and description should be unchanged.
Confirm the edited file is .agents/skills/exec-summary/SKILL.md. Editing a loose extracted copy elsewhere would not update the installed skill.
Start another fresh chat for testing. If the old rule still appears to be loaded, fully quit and reopen Codex before testing again.

Read the exact edit prompt

Open .agents/skills/exec-summary/SKILL.md in this practice folder.
Replace only the line "- 120 words maximum in total. Count before you answer." with "- 60 words maximum in total. Count before you answer."
Leave the name, description and all other instructions unchanged. Show me the before-and-after line and a diff. Do not generate a summary yet.

Checkpoint The discovered skill file contains the 60-word rule. You checked the edit before running the task.

6. Run the task again and check 60 words

In the fresh chat, keep the same model and reasoning setting.
Copy and send the same Pedala prompt again. Save the answer as Codex — Edited skill.
Count all words, including labels. Use a word-count tool or the count-only prompt below as a quick check; verify any model-provided count.
Compare this answer with the original with-skill answer. Record whether it meets 60 words and what changed.

Read the exact pedala prompt

Summarise these notes for my manager:

Call notes– Pedala (e-bike subscriptions, Barcelona) – Tuesday
- attendees: Marta (CEO), Jordi (ops), us
- ~1,200 active subscribers; monthly churn up from 4% to 7% since June!!
- Jordi thinks it's repair wait times: avg 6 days now vs 2 days in spring. Their mechanic left in May
- Marta more worried about new competitor BiciYa, launched at €29/month vs Pedala €39/month
- exit survey: 140 responses, 58% of churned users mention repairs, 22% mention price
- runway ~9 months
- options discussed: hire 2 mechanics (~€6k/month total), cut price to €32, or partner with a repair shop in Gràcia (no quote yet)
- Marta wants a decision before the board meeting on 15 Nov
- we owe them a recommendation by next Friday
- someone mentioned Valencia expansion but parked

(Pedala and BiciYa are fictional companies made up for this exercise.)

Read the exact word-count check

Count the words in the assistant's summary immediately above, including its labels. Report the count only. Do not rewrite the summary.

Checkpoint You have a third answer tied to a reviewed one-line edit, rather than a follow-up request that merely asked for a shorter summary.

7. Test automatic activation

Leave the edited exec-summary installed. Start a New chat in Skills Workshop, copy the Peru prompt, and send. Prediction: No.
Start another New chat in the same folder. Copy the article prompt including the article, then send. Prediction: Yes.
Start a third New chat in the same folder. Copy the poem prompt and send. Prediction: No.
For each run, inspect the visible activity if available and record Yes, No or Unknown. Do not add $exec-summary to these prompts: that would force selection.

Read the exact peru test

What's the capital of Peru?

Read the exact article test

Can you condense this article into something my boss can read in 30 seconds?

Campus bike repair pilot

The student services team tested a bike repair desk for 4 weeks. It served 80 students, and 60 booked a second visit. Average turnaround was 2 days, compared with 5 days at the previous provider. The pilot cost €1,600. The team has a €2,000 budget for next month. It has not yet checked whether the provider can handle twice as many bookings. Ana will collect a capacity quote by Friday. The committee must decide on Monday whether to extend the pilot for another month.

Read the exact poem test

Write a poem about Barcelona.

Checkpoint Three fresh chats test the description’s boundaries. A matching-looking output alone does not prove that a file was read.

Keep the evidence.

Score each check with 0 or 1. Count the whole output, including labels. Leave uncertain checks blank until you verify them. Keep the AI answers in your own notes.

Claude: one point for each yes
Check	Without	With
Opens with a recommendation, not the topic
Exactly 3 reasons; each has a number from the notes
120 words or fewer, including labels
No invented numbers; every number traces to the notes
Ends with who does what by when (TBD is allowed for missing details)
Total	— / 5	— / 5

Codex: one point for each yes
Check	Without	With
Opens with a recommendation, not the topic
Exactly 3 reasons; each has a number from the notes
120 words or fewer, including labels
No invented numbers; every number traces to the notes
Ends with who does what by when (TBD is allowed for missing details)
Total	— / 5	— / 5

Did the skill activate?

Claude: activation observations
Test	Expected	Observed file read
Peru	No
Article	Yes
Poem	No

Codex: activation observations
Test	Expected	Observed file read
Peru	No
Article	Yes
Poem	No

Claude: what changed after the edit?

Codex: what changed after the edit?

The initial comparison uses a 120-word limit. For the edited version, check 60 words separately. A tie is allowed; fluent formatting does not establish factual accuracy.

Optional · try the STAR demo yourself

Do this after the core exercise, or follow the facilitator’s demo. Keep the two workshop skills from competing.

Claude

In Customize → Skills, turn exec-summary off.
Upload star-feedback.zip using the same upload steps, and enable it.
Open a new chat, paste the interview prompt below and send.

Codex

In a setup chat in Skills Workshop, send the STAR-install prompt below. Review the moved and installed folders.
Open a fresh chat in that same folder; restart the app if discovery has not updated.
Paste the interview prompt and send. If needed, try an explicit $star-feedback request in a separate chat.

Read STAR installation prompt

In this practice folder only, move .agents/skills/exec-summary to parked-skills/exec-summary so it is outside the discovered skills directory. Keep that copy; do not delete it.
Extract star-feedback.zip and install its star-feedback folder at .agents/skills/star-feedback. Keep references/rubric.md inside that folder. Stop if a destination already exists. Show the resulting tree. Do not score an interview answer yet.

Read the interview prompt

Tell me about a time you led a team. Here's my answer:

In my previous job at a consulting firm we had a project that was delayed. The client was unhappy and the team was stressed. We decided to have daily stand-ups and I talked to the client a lot. In the end the project was delivered and the client was happy. I learned that communication is really important.

Checkpoint Look for reads of SKILL.md and references/rubric.md when visible, four scores from 1–4, and a rewritten answer that uses placeholders instead of inventing achievements.

If something does not match

Claude: I cannot find Skills or upload a ZIP

Check code execution under Settings → Capabilities, then go to Customize → Skills. Refresh the page after enabling it. If an organization manages the account, its owner may need to allow code execution and skills. Ask a helper or pair up if blocked. Do not buy a plan just to follow this guide; Free accounts are supported.

Claude: my edited skill still gives 120 words

Reopen the file inside exec-summary-60.zip and check it contains 60. Confirm you uploaded that ZIP, removed the original workshop entry and enabled the new one. Run in a new chat. If it still fails, record a failed test; do not assume a correct upload guarantees compliance.

ZIP: I see two exec-summary folders

Find the inner folder that directly contains SKILL.md. That is the folder to compress or install. The final Claude ZIP path is exec-summary/SKILL.md; the final Codex path is .agents/skills/exec-summary/SKILL.md.

Codex: I cannot see .agents in Finder or File Explorer

Mac: press ⌘Shift. (Command, Shift, period) to show hidden items. Windows: File Explorer → View → Show → Hidden items. Or open the file link Codex gives you. The dot folder is normal.

Codex: the local skill is missing or two copies appear

Confirm the chat is in Skills Workshop and the file is at .agents/skills/exec-summary/SKILL.md. Restart Codex if needed. User-wide or plugin copies can coexist with the local skill; ask a helper to disable the other workshop copy. Do not delete unrelated skills or replace your entire config file.

Codex: Git initialization is blocked

The project-local route uses a practice Git repository so discovery has a clear root. If the app reports Git is unavailable, ask a helper to install/enable Git or use the Claude route. Do not change to your work repository to get past this step.

I cannot tell whether the skill was used

Record Unknown. Visible style and the model saying “I used it” are weaker evidence than a file-read trace. For troubleshooting, explicitly request the named skill in a separate fresh chat; keep that result separate from the automatic trigger tests.

The without-skill answer scores just as well

Keep the result. The model can already produce good summaries. Compare whether the requested format is consistent and then test the edited 60-word rule. A tie is useful evidence, not a reason to alter the baseline after seeing the answer.

The Copy button does not work

Open the matching “Read the exact…” panel. Select the text inside the prompt box, copy with ⌘C or Ctrl+C, then paste into your app. This also works when the browser restricts clipboard access on local files.

Optional: I already use Codex CLI

Use the same local folder and prompts. Open Terminal or PowerShell, type cd (including the space), drag the Skills Workshop folder into the window, then press Enter. Type codex and press Enter. Sign in if asked. Use /new for each fresh conversation; if your version does not offer it, exit and relaunch codex in the same folder. Use /skills to inspect the selector or $exec-summary for a separate forced test. This optional route assumes the CLI is already installed; it is not required for the workshop.

Sources and scope

Product steps checked 15 October 2026. Exact interface labels can vary by version. The exercise design and prompts are workshop material; product installation and loading behavior are grounded in the sources below.

Only complete one route in class. The separate setup and edit conversations prevent instructions from those conversations becoming part of the measured summary chat. Fresh chats still share account-level preferences and installed skills; this is a practical classroom comparison.