A collection of 12 skills extracted from a working Salesforce delivery repo and de-identified for public use: 7 on Salesforce platform knowledge, 2 on Jira, and 3 working methods that are not platform-specific.
| Skill | What it does |
|---|---|
salesforce-flow-and-ui-defects |
Catalogs the Flow and Lightning failures that pass a clean deploy and surface only on screen, with the cause and the fix for each: the 15-against-18-character $User.Id comparison, a blank cross-object merge field, a picklist default that reaches no existing record, an LWC missing from phones, a layout-fed related-list tab, an orchestration Resource that wants a Username, and a validation rule that rolls back the record it guards. Also covers how to exercise a screen flow whose Submit is destructive. |
salesforce-flow-versions |
Establishes which version of a flow an org is actually running, since source status, a deploy result, a retrieve, and FlowDefinitionView each report on a different one. Covers activation and deactivation, the three routes to an InvalidDraft that deployed green, deleting a flow safely, why a dependency query returning nothing does not mean a field is safe to delete, and why a running orchestration keeps the stages it started on. |
salesforce-data-loading |
Walks the design of a bulk load: which describe flags decide the approach before you build the file, the two permissions that let a load set audit fields and inactive owners, how to derive an undecided value from records that already carry it, and the four operations that report success while doing something other than what you meant. |
salesforce-user-automation |
Explains why automation touching User alongside any non-setup object breaks, and how to structure it so it does not: the async path, the chained Queueable, and tests that actually exercise the collision. Also covers User fields that describe as nonexistent, and querying Chatter membership without silently dropping inactive members. |
salesforce-business-days |
Sets up Business Hours properly: how to measure what a set really is, since the metadata cannot tell you, a two-set design that lets holiday-aware and holiday-blind automation coexist, the deploy semantics for holidays, and when business-day math needs Apex rather than a plain TODAY() + 7. |
salesforce-code-analyzer-gate |
Diagnoses a red Code Analyzer check: which rules can fail the job rather than report, what ApexCRUDViolation is asking for, when a finding comes from your branch's config vintage rather than your code, and how to force a check to re-run without dismissing an approval. Includes the Windows locale fix for the flow engine. |
salesforce-files-queries |
Builds a file population from ContentDocumentLink rather than a ContentVersion sweep, explains why the obvious sanity check confirms the wrong number, and says where to chunk a long IN list. |
| Skill | What it does |
|---|---|
jira-rest-writes |
REST recipes for every Jira Cloud write, each paired with the read-back that confirms it: create, edit, comment, transition, assign, link, sprint, and Story Points. Covers building ADF documents, resolving an assignable account on a site that also runs Service Management, and which fields have to go in a call of their own. Ships with a converter that turns plain text into the ADF payload the endpoints expect, so paragraphs, bullets, and numbered lists survive instead of flattening into a single block. |
qa-handoff |
Writes a QA sub-task a tester can act on without you: numbered one-check-per-step guidance, the field combination that drives the behavior rather than fixture record Ids that drift, screenshots placed at the step they illustrate, desktop and phone captures, and the order to create and transition sub-tasks in so Jira automation does not close the parent out from under you. |
| Skill | What it does |
|---|---|
design-grilling |
Works the open decisions on a piece of work as a tree, in rounds, putting the whole frontier at once with a recommended answer to each, and separates the facts you should look up yourself from the decisions only a person can make. |
session-handoff |
A four-section handoff file per task, and the rule for when to read and write one, so a fresh session continues the work rather than re-deriving it. Carries the cost case for actually stopping: re-read context is billed every turn, so it compounds with position, and in one team's measured usage a request past turn 200 re-read 619K tokens against 24.8K at turn 1, with 47.7% of all spend sitting past turn 100. |
stakeholder-questionnaire |
Turns questions you cannot answer into one document a named colleague fills in, by interviewing whoever is directing the work about the send rather than about the subject. |
Copy the ones you want into your skills directory. Personal, available in every project:
cp -r skills/salesforce-user-automation ~/.claude/skills/Or per project, checked in with the repo, so the whole team gets them:
cp -r skills/design-grilling .claude/skills/Claude Code loads each skill's description on every request and the body only when the description
matches what you are doing, so install the ones you will use rather than all 12 out of habit.
These were written for day-to-day Salesforce delivery work and rewritten for general use, with names, identifiers, and organization-specific references removed.
What stayed is the mechanism and the evidence for it. Where a claim rests on a single measurement it
says so, and where a cause was never established it says that too rather than offering a tidy
explanation. The measurements come from real environments during 2026, but all from one organization,
so verify anything you plan to rely on against your own. And design-grilling closes with a list of
Salesforce decisions drawn from that same source, which the skill tells you to replace with your own.
MIT. Three of the 12 are adaptations of Matt Pocock's MIT-licensed skills; see NOTICE.md for what came from where and what changed.