# Use the kit in your existing AI assistant

The portable path is the same: attach labeled sources, the operating brief, `prompts/session.md`, the chosen prompt pack and `skills/practicalworkai/references/records.md`. Run one stage at a time. File creation and code execution depend on your account and tool availability; complete fenced CSV is the fallback. This kit does not connect to an API or need a key.

## ChatGPT
Use a fresh chat for the first practice run. For repeated work, create a Project if available, add the session instruction as project instructions, and add the relevant source files and record schema. Keep each case's authority and operating brief explicit. Ask for actual CSV files where file creation is available; otherwise save the returned CSV blocks. Verify whether code was actually used for calculations.

Official setup reference, checked 5 October 2026: [ChatGPT Projects](https://learn.chatgpt.com/docs/projects?surface=app). On that date, three small synthetic workflows were executed through a signed-in ChatGPT UI in Chat mode at the displayed Medium setting. The exact model identifier was not exposed. The WBS exported successfully; one date variant and the proposal records needed external review corrections. See VALIDATION.md for the tested inputs and limits. This does not establish availability or reliability across accounts, models or Projects.

If ChatGPT offers to switch to Work, choose the mode you intend to evaluate; the walkthrough stayed in Chat. File export can take time and encounter errors. Verify the downloaded contents independently. If a file download is inaccessible, ask for complete fenced CSV/text and save it without silently repairing values. Run the kit's actual offline checker against the source files and exported records; a model's self-written checker is a separate check. The portable checker was not attached during the initial walkthrough.

## Claude
Use ordinary chat with the same source and prompt packs. Where custom Skills are enabled, use the site's separate `practicalworkai-skill.zip` in Claude's Skills upload interface. The ZIP contains one `practicalworkai/` folder with SKILL.md, references and a dependency-free checker. Enable it and ask for the selected workflow; attach your case sources separately. Check your account's file/code capabilities rather than assuming the checker ran.

Official reference: [Use Skills in Claude](https://support.claude.com/en/articles/12512180-use-skills-in-claude). Packaging is structurally checked; a Claude import and full run remain untested.

## Gemini
Use chat for the first case. For reuse, create a Gem where available, put the session instruction and desired workflow into Instructions, and add the workflow pack/schema under Knowledge. Save the Gem; preview alone does not save it. Add case-specific sources and decision context to each task. Export CSV blocks if file creation or code execution is unavailable.

Official reference: [Create and use Gems](https://support.google.com/gemini/answer/15235603?hl=en). This describes a supported setup route; it is not evidence of a completed Gemini run.

## File-capable agents
Point the agent at `skills/practicalworkai/SKILL.md` or install that folder with its documented skill mechanism. Supply an input folder, output folder and operating brief. Ask it to read the selected reference, save records, run `scripts/check_records.py` and return the brief. The skill should not be installed into a global folder without your intent. It needs no network access to run the checks.

One independent agent run per core case has been reviewed for this edition (see VALIDATION.md). Those narrow synthetic tests do not establish cross-service compatibility or reliably better professional outcomes.
