# Optional tool guide

Documentation checked on 14 October 2026. These instructions were researched from official sources; the course author did not execute them in a logged-in Gemini or Make account. Interface labels, account eligibility and quotas can change. Use an existing account only if the feature is available and its use is permitted. Stop if the route asks for a purchase, new account, API key, real data or an unapproved permission change.

## Choose a route

The core learning uses the supplied synthetic files and workbook. You can complete a design and rehearsal with no account. Ordinary existing chat can run the portable configuration manually. A saved-assistant feature makes that configuration reusable across conversations. The optional Make lab demonstrates no-code data movement and routing. The assistant and workflow are joined by a declared human validation step, not an unbuilt AI API integration.

## Saved assistant using Gemini

Google's current Skills documentation describes a gradual rollout for personal accounts, age 18 or over, with Keep Activity enabled. Skills do not require a Google AI subscription; scheduled task features have separate eligibility. If these conditions are not already met or the feature is absent, use the fallback rather than changing privacy settings for this course. [S2]

Where available, open Gemini's Customize area, then Skills, then Create manually. Enter a descriptive name, the purpose and your Module 2 instructions. Create it, then explicitly request it in a new chat with the synthetic policy and request. Save the output and compare it with your expected result. The course uses pasted reference text, so no file upload is necessary for this path. Google also supports other creation routes; they are outside this short lab. [S2]

If your account instead shows Gems, the documented web route is Explore Gems, then New Gem. Add a descriptive assistant name and instructions, test in the preview and save. Preview alone does not save the Gem. Knowledge can hold supplied reference files where supported. For this course, paste the short pack into the conversation or upload only the synthetic active file if you choose; do not connect Drive. Keep citation visibility enabled. Google lists Gems without a Google AI plan, but availability and usage limits still apply. [S3, S4]

Do not enable connected apps, automatic schedules or public sharing. A saved configuration does not need those capabilities to complete the course. Test R101, R104 and R105, then repeat in a fresh conversation. Record whether your result came from a saved feature or an ordinary chat.

## Ordinary chat or offline fallback

Paste the stable configuration, the active synthetic pack and one request into an existing permitted chat. Save them as separate local files so you can replace the pack without rewriting the instructions. Mark this as a manual reusable prompt. If you have no model access, use the workbook's authored illustrations and trace the same decision process. Mark those results manual rehearsal. Neither fallback requires account creation or spending.

## Make lab A Map one record

Make's built-in tools can operate without a third-party app connection. Its current documentation allows a scenario to begin with a non-trigger module and be run manually. [S5]

In an existing permitted Make account, create a training scenario named Harbor synthetic routing. Keep scheduling off. Add Tools → Set variables (older labels may say Set multiple variables). Add request_id R101, revision 1, item projector, duration_days 2, final_route DRAFT_READY and draft_text “Two-day projector request awaits desk review.” Use one-execution lifetime where offered. Save and use Run once. Inspect the output bundle to confirm the values, then retain a text record of what appeared. Scenario variables last within a run; they are not a persistent review queue. [S6]

These are prevalidated inputs for a wiring lab. The human performs the Module 5 checks before entering final_route. The scenario does not validate the full policy or interpret untrusted messages automatically. Label those checks manual in your workflow map.

## Make lab B Route three records

Attach a Router to the first module. Create three branches. The first is named Standard draft, the second Missing information, and the third Hold for review. Make processes routes sequentially; more than one matching route can still process data. Use exact, non-overlapping route values. Mark the third as the fallback for unmatched values. [S7]

Set the first branch's filter to final_route exactly equal to DRAFT_READY and the second to final_route exactly equal to NEEDS_INFO. Filters are configured on the connection between modules and pass only matching bundles. Use the mapped final_route token from the first module, not a hard-coded copy of its current text. [S8]

At each branch's end add Tools → Set variable. Use a variable named result_summary and a value containing the mapped request_id, original final_route and a brief label: standard review pending, missing information pending, or hold with original reason. These are visible run outputs, not durable records.

Run three inputs separately: R101 with DRAFT_READY; R104 with NEEDS_INFO; R102 with HUMAN_REVIEW. Expect one branch per run. The fallback must preserve HUMAN_REVIEW in its displayed result. This wiring lab displays results only; a duplicate or technical hold must never create an additional queue item in an extended design. Also test an unknown value TYPO; it must land on hold. A blank route should be held too. Record the actual branch and output each time. A successful run symbol is insufficient evidence.

If filters accidentally match both branches, repair them and rerun all cases. Do not add email, messaging, booking or AI API modules. At the end, leave scheduling off and retain the scenario as a clearly labelled synthetic draft.

## Optional persistent queue extension

This extension is additional practice after Module 6, not required for the base lab. Use only an unused permitted training Data store; do not change an existing shared store or buy capacity.

Make documents a one-store, 1 MB allowance on its Free plan. A Data store persists across runs. Create a training structure with request_id, revision, final_route, draft_version and review_state. Use a stable key such as R101|1|queue_review. With Add/Replace a Record, explicitly disable overwriting. The documentation says adding an existing key then raises an error. Inspect the store after the first write and after replay; confirm there is still one record. Keep the duplicate error visible for reconciliation, rather than deleting or overwriting the existing item. [S9]

This test checks the documented keyed-write behaviour in your account. It does not establish an authenticated multi-user approval system or prove every concurrent/retry failure path. Record the limitation and stop on any unexpected error. Do not test with a production destination. No deletion exercise is required; follow your permitted retention process later.

## Cost and error handling

Make currently lists 1,000 monthly credits on Free. Credit consumption varies with the modules and AI features used. Check remaining capacity before and after the small lab; keep the actual usage evidence. Do not assume that one run equals one credit. Paid capacity, AI providers and third-party services are separate considerations. [S10]

Make's incomplete-execution storage is documented as disabled by default. Its recovery features depend on settings and supported error types. Inspect the documentation and your settings before claiming a retry or retained failed run exists. The course's failure cards are simulations unless you separately observe the relevant behaviour. [S11]

## Why custom GPT creation is not required

Current OpenAI documentation distinguishes personal accounts from managed workspaces: personal Free, Go, Plus and Pro accounts cannot create or publish new GPTs, while Business, Enterprise and Edu creation depends on workspace permissions. Existing GPTs have separate editing eligibility, and the documentation describes retirement/migration plans. The course therefore uses a portable configuration rather than making GPT creation an entry requirement. Recheck the official page before choosing that optional provider. [S1]

## Official sources

S1 OpenAI Creating and editing GPTs: https://help.openai.com/en/articles/8554397-creating-and-editing-gpts
S2 Google Create and manage skills for Gemini Apps: https://support.google.com/gemini/answer/17094296?hl=en
S3 Google Tips for creating custom Gems: https://support.google.com/gemini/answer/15235603
S4 Google Gemini Apps limits and upgrades: https://support.google.com/gemini/answer/16275805?hl=en
S5 Make Types of modules: https://help.make.com/types-of-modules
S6 Make Scenario variables: https://help.make.com/scenario-variables
S7 Make Router: https://help.make.com/router
S8 Make Filtering: https://help.make.com/filtering
S9 Make Data stores: https://help.make.com/l6du-data-stores
S10 Make Pricing: https://www.make.com/pricing
S11 Make Incomplete executions: https://help.make.com/incomplete-executions