1Introduction

This document is a translation in substance. The reference is the original document in French; extensions and changes are always made there.

This guide is the session tool. It does not repeat the presentation of Grok: it says what to do, in what order, with which messages. Each section stands on its own and can be consulted mid-work.

Audience: users of the Kit and of consumer projects who work with Grok. Tone: direct.

2Session card

# Habit Effect
1 One goal per session Keeps conversations from sprawling
2 Outline first, content after Fixing a structure costs less
3 Artifacts are the source of truth Chat memory does not survive on its own
4 Announce, then wait for the go-ahead You keep control of the files
5 Strict stylesheet levels Readability and validation
6 Validate before delivering kit_​validate_​docx exit 0
7 Correct in a targeted way Problem, direction, what to keep
8 Say what you refuse as well Less filler
9 Close formally Summary, inventory, open points
10 Full Kit injection No partial update

3Running a session

3.1Opening

The first message sets the frame: project, goal for the day, files to reread, constraints of language or format. Ask Grok to confirm what it has read — the stylesheet version in particular — before anything is produced.

New session on [PROJECT NAME].
Goal: [PRECISE GOAL].
Read the artifacts, in particular [KEY FILES].
Confirm the stylesheet version you have read.
Announce the plan. Run nothing until I have said go.

3.2During the work

One section at a time. Approve it before moving to the next.

Files in the project. Point to the artifacts rather than pasting long extracts again.

A short reminder. If a rule drifts, restate it in one sentence. Do not explain everything again.

Open points. Note what comes up and is not being dealt with now.

3.3Closing

Five minutes of closing save you explaining everything again in the next session. The Kit's short formula is enough.

And now for the end of the session, please.

Expected: files produced with their exact names and timestamps, decisions, open points, suggestions for what follows, files to place in the artifacts.

4Wording

4.1Three questions before sending

What. Is the deliverable named (outline, section, script, document)?

For whom. Are the audience and the level stated?

Under what constraints. Language, length, format, what not to do?

If any of the three answers is vague, complete the message before sending it.

4.2Show rather than describe

One example of input and output aligns the format faster than a description. Useful for extractions, rewrites and repetitive tables.

Format wanted, one line per item:
Name | Type | Location | Value
Example: Living room temperature | Temperature |
         Living room | 21.3 °C
Now process the attached file.

4.3Correcting without breaking everything

Regenerating without changing the instruction produces a result of the same kind. If the prompt was imprecise, write a corrective message.

The section [NAME] does not work:
- [PROBLEM 1]
- [PROBLEM 2]
Rewrite it with: [DIRECTION].
Keep: [WHAT TO PRESERVE].
Note:Avoid “this isn't good, do it again”. Say whether it is the substance, the structure or the tone, and give the direction.

5Standard workflows

5.1A professional document

# Action Stop while
1 Framing: audience, scope, language The framing is not approved
2 Detailed outline of the sections The outline is not approved
3 Writing section by section The current section is not approved
4 Stylesheet script and setters The upstream checks fail
5 Generation and kit_​validate_​docx The exit code is not 0
6 Delivery into the artifacts The name or the timestamp is wrong
💡Beyond a dozen sections, one session per group of three. The context window is not a binder.

5.2Analysis and diagnosis

Attach every useful file with the first message.

Ask for step-by-step reasoning. Inventory, causes ruled out, diagnosis, then actions.

Approve the diagnosis before the recommendations.

Close with a synthesis worth filing.

5.3A script or an automation

Specify the inputs and the outputs before a line is written. Ask first for the list of stylesheet functions to call. Attach the active stylesheet. Test each block before moving on.

For a Kit script: the Reference read within the current session, contracts respected, no call outside a helper.

6Further templates

6.1Picking up after a break

I'm picking [PROJECT NAME] back up after [HOW LONG OR CONTEXT].
Read [KEY FILES].
Summarise the current state in five points.
Goal for this session: [GOAL].

6.2Generation once the content is approved

The content of sections [LIST] is approved.
Generate the full script.
Stylesheet: [FILE + VERSION]
Output name: [PREFIX] - [CATEGORY] - [SUBJECT], timestamp
             via getLuxTimestamp()
Author: [NAME]
Call the Registry setters, then the full check chain before
any delivery.

6.3Analysing a technical file

Here is [FILE TYPE]. [ATTACHED FILE]
Produce:
1. A structured inventory of the [ITEMS].
2. Anomalies or points of concern.
3. [OPTIONAL: recommendations].
Reason step by step before concluding.
Note:These templates are starting points. Precise wording beats a template followed word for word.

7Costly mistakes

Do Avoid
One clear goal per session Mixing everything into one long conversation
Approve the outline before the content Asking for a complete document in one go
Keep the artifacts current Relying on chat memory alone
Stylesheet helpers at the right levels Manual formatting outside the styles
Announce, then go Letting generation start without an approved plan
Validate the .docx before delivery Delivering an unchecked file
A note or a tip after the useful content A note or a tip right under a heading
Full Kit injection Copying three files and ignoring the rest