Training Impact

Training impact

Three repositories move together now: sfdx-hardis, vscode-sfdx-hardis and sfdx-hardis-training. The existing rule that every change states its VS Code extension impact has a sibling: every change states its training impact, even when it is "none".

This skill answers one question: which labs does this change break, and does it need a training Pull Request?

It never edits the training repository. [[training-update]] does that, once this says there is an impact.

The sibling layout

sfdx-hardis-training is always a sibling of this repository, exactly like vscode-sfdx-hardis already is:

C:/git/  (or wherever this repository lives)
├── sfdx-hardis/            <- the skills live here, and only here
├── sfdx-hardis-training/   <- sibling, cloned if absent
└── vscode-sfdx-hardis/     <- sibling, the screenshot harness

The path is never configurable: a fixed layout is what lets the skills act without asking.

Clone it if it is missing, do not fail and do not work from a temporary directory:

[ -d ../sfdx-hardis-training ] || git clone https://github.com/hardisgroupcom/sfdx-hardis-training.git ../sfdx-hardis-training
git -C ../sfdx-hardis-training fetch --all --prune

Read from the latest main unless the user says otherwise.

What to read

../sfdx-hardis-training/training-manifest.json, generated from the depends_on front matter of every lab. It holds, per lab: the commands, flags, config keys, documentation pages, panels and screenshots it relies on, plus a reverseIndex keyed by each of those.

node -e "
const m = require('../sfdx-hardis-training/training-manifest.json');
const key = process.argv[1], value = process.argv[2];
console.log((m.reverseIndex[key] || {})[value] || 'no lab depends on it');
" commands hardis:work:save

It works offline and against the exact checked-out state, which is what makes this a check rather than a habit.

The course is in two languages, and one of them is the reference

The training ships in English and in French: labs/en/ and labs/fr/, mirrored file for file.

labs/en/ is the reference, and the impact is always computed from it. The manifest is generated from the English front matter, so reverseIndex names English lab ids and nothing else. That is correct and it is not a gap: a lab's depends_on is the same in every language, because the commands, flags, config keys and panels it relies on are the same.

What it means for the answer you give: a lab that is affected is affected in every locale. Each manifest entry carries a translations block naming the file and the URL of that lab in each other locale, so name them:

node -e "
const m = require('../sfdx-hardis-training/training-manifest.json');
const lab = m.labs.find((one) => one.id === process.argv[1]);
console.log(lab.file, '+', Object.values(lab.translations || {}).map((t) => t.file).join(', '));
" lab-2-3

A change never breaks only the French version, and a French-only fix is never the answer: the correction goes into labs/en/ first and the translations follow. [[training-update]] has the procedure.

The procedure

1. List what the change touches. Be specific and literal:
   • command ids (hardis:work:save)
   • flags (--check, --agent)
   • config keys (autoCleanTypes, useDeltaDeployment)
   • documentation page slugs (salesforce-devops-config-cleaning)
   • LWC panels (pipeline, backpromote, deploymentAction)
   • report file names and shapes
2. Look each one up in the reverse index.
3. Classify the impact with the table below.
4. State it in the analysis and in the design, even when it is "none". Name the labs.

Classifying

The screenshot rows are never just a re-capture. Every click a lab describes has to be shown, the pills have to be redrawn and every image has to be looked at again. The hard rules for that live in [[training-update]], section Screenshots: read them before sizing a panel change.

That last row is the dangerous one. The training deliberately seeds failures (a field excluded by .forceignore, a field that cannot be made required, a hardcoded id that trips PMD). A change that makes one of them stop failing turns a lab into a page describing something that does not happen, and no test anywhere notices.