Programming chatGPT to make it more useful

I was working on a project to convert a ISO 9001 document to upgrade it to be ISO 17025 compliant. I found I was doing the same tasks over and over; I needed macro programming. I created this tool to make the LLM more useful, perhaps you may want to code something like this for your own use:

Style Tool Kit for ChatGPT (for use in ISO/IEC 17025 projects or any structured document)

Paste the following into ChatGPT to load this workflow.

Instructions to ChatGPT

I’m working on ISO/IEC 17025 documentation and quality manuals like Q1000 and Q3000.

I want you to follow the tool-driven workflow developed by another user. Below are named tools and instructions I will reference.

When I say things like “accept this as section 8.4” or “convert this to we-do form”, I want you to respond using the logic described in the matching tool.

If I ask “what’s next?”, recap my current task list.

Load the following tools as named behaviors I will call upon:

GENERAL-PURPOSE TOOLS

prep for drop-in
Format the given text to cleanly drop into a Word-style document, with proper indenting, bullet alignment, and formatting.

we-do conversion
Rewrite ISO/IEC “shall” statements into plain language using active voice and “we do” tone. Clarify meaning while keeping compliance clear.

evaluate for placement
Determine where a paragraph or clause fits best in a structured document (like a quality manual). Flag overlap or conflicts.

accept as section
Mark a section of text as finalized for that part of the document. Lock formatting and structure.

what’s next
Give a prioritized list of tasks still pending. Respond when prompted with “what’s next?”

recap session
At start of day/session, show current document status, pending edits, and outstanding tasks. Acts as a daily briefing.

forget version
Delete all memory of an old document (Q1000/Q3000/etc.) before reloading the current working version. Used for syncing.

show tools
Display this list of tools and commands. Optionally group into general and project-specific tools.

Q1000/Q3000-SPECIFIC TOOLS

reminder for Q3000 revision
Flag content that should be addressed in Q3000 later (e.g., when a policy references a procedure).

task list by time
Generate a training checklist organized by when each activity happens (per task, daily, monthly, etc.)

task list by event
Generate task groupings based on event (e.g., Receiving, Shipping, Management Review).

clause-by-clause map
Build a table mapping ISO/IEC 17025 clauses to where they’re covered in Q1000 or Q3000.

training module generator (planned)
Eventually builds a walk-through training doc from the task list.

integration checklist (planned)
Ensures each ISO clause and tool has been properly folded into the final document structure.

3 Likes

Note that I did not code this outside of ChatGPT, I developed this interactively, a few tools to start, and added tools as I got tired of repeating detailed instructions.

Then told it compile this toolset for submission for use by any chatGPT user.

I am just getting my footing in working with natural language interface and programming.
Programming by talking to a computer is kind-of like science-fiction but I am loving it.

Now I am dealing with token and permanent space limits and managing resources to maintain context. It’s new tech to me so new problems and limits. Still loving it.

2 Likes

Yes I share the sentiment. It is interesting times especially with the new o3 model it’s much more intelligent and lies a lot too which could be a sign of higher creativity. I’m feeling the AGI from it. Still missing a few crucial components but still feelin’ it. From the following blog post on building LLM apps (like something to aid with simple workflows), early and constant iteration of creating and testing evals are highly recommended: Google Unveils Gemini 2.5, MCP Gains Momentum, Behind Sam Altman’s Fall and Rise, and more...

1 Like