agentleFS
omoji-personalomoji-personal/torque/.agents/skills/salesforce-architecture-review/SKILL.md

salesforce-architecture-review

Skill0 starsChanged 3 days agoMITView on GitHub

Review a Salesforce data model, automation design, integration, access model or cross-component solution against business outcomes and maintainability. Use it when asked to review, assess or sign off a design or an existing org, to compare design options, to judge whether a build is over-engineered or risky, or before a consequential design decision is recorded. Gives a review method, the five Well-Architected pillars as a lens, a checklist of forty rules by phase, and a catalogue of forty anti-patterns with how to detect each.

The file

We split this skill into the rules it gives an agent

1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.

  • ≈0

    Common · estimate

    Many files like this have a rule of this kind.

  • ≈1

    Distinctive · estimate

    This author's own choices. Worth a look.

  • 0

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Distinctive:

    Do not assert a platform limit from memory: cite it or mark it to check.

    line 65

≈ “Common” is an estimate: at least 1 in 5 of the 372,429 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

Salesforce architecture review

A review reads and reports. It does not change the org or prove that an implementation works. State described: Winter '27 (API 68.0), checked 2026-11-14. To build, use skills salesforce-build and salesforce-flow; the decision tables live there.

Marks: verify = not confirmed in a fetched primary source: verify in the org or the current documentation. "Lead" = a query or element name written from the research, not run for this release: confirm names with describe, and treat an error as "not checked".

Method

  1. Fix the question. What outcome must the design serve, for whom, and what decision does the review feed? Write it in two lines and get a yes.
  2. Gather evidence, read-only. The design document or the metadata; the org's shape (product family and installed packages, automation already on the objects in scope, the sharing model, data volumes); who will maintain it. Name the org and the date on each fact.
  3. Separate four things as you read: what a managed package does, what the client customised, what the client decided, and what is assumed without evidence.
  4. Trace one real transaction end to end: the page a user opens, the access that lets them, the automation that fires, the data written, the outside systems reached, what happens on failure and on the second attempt.
  5. Check it against the anti-pattern catalogue (references/anti-patterns.md) and the rules for its phase (references/review-checklist.md). Apply only what the change needs.
  6. Weigh trade-offs with the five pillars (references/decision-methods.md). Do not maximise every pillar: say which trade-off was made and whether it fits the size of the org.
  7. Look for the simpler design. Would a standard feature, an installed package, or configuration do the job? Is the effort in line with what the client was told? A simpler native option or an over-budget plan is a top finding, not a footnote.
  8. Report (below), and state what still needs confirmation.

The five pillars, as questions

PillarAsk
TrustIs access least privilege by default? One identity per integration? No secrets in code or configuration? Data classified at field level?
ReliabilityDoes it hold at 200 records and at the real volume? What happens when a step fails, and on a retry? Can the data be restored?
Operational ExcellenceCan it be deployed, validated and rolled back from source? Who hears about a failure? Is it documented where an admin will look?
Resource and Cost OptimizationWas what the org already owns weighed first? Who maintains it, at what cost over three to five years? Any unused automation left behind?
FairnessCan everyone use the pages (labels, keyboard, contrast)? Does a rule or a required field shut out people who lack the data? Is a human in the loop for high-stakes decisions?

Source: [WA-FW]. The older Trusted, Easy, Adaptable structure was replaced on 2026-09-14 and its page addresses no longer resolve [WA-BLOG]: treat any guidance that cites it as stale.

What to report

FieldContent
SeverityBlocks the decision (data loss, exposure, cannot work at volume, cannot be undone) / Fix before production / Improve when next touched
FindingOne sentence, on one component or decision
EvidenceThe file and line, the query and its result, or the document section; org and date
Why it mattersThe outcome or the maintenance cost it affects, in the client's terms
RecommendationThe smallest maintainable change; the trade-off only when it changes the decision
ConfidenceRead in source, observed in the org, or assumed
  • Report one defect once. Treat untested behaviour as a gap, not a defect.
  • Do not assert a platform limit from memory: cite it or mark it to check.
  • Say what you checked and what you did not.
  • A design review is not evidence that the implementation works. When implementation is also requested, continue with the agreed design and the build skills.

Stop and ask a person when

A change trades one pillar against another (speed against access control, cost against recovery); a target such as availability, recovery time, retention or a compliance figure would have to be invented; the framework and the client's own convention disagree; the design touches anything on the stop-and-ask lists in references/review-checklist.md.

References

  • references/review-checklist.md: forty rules by phase, each with where the detail lives.
  • references/anti-patterns.md: forty smells, how to detect each, the better pattern.
  • references/decision-methods.md: pillars and tensions, proportion, options and decision records, org strategy, operations, what changed recently.

A path that starts with another skill's name, such as salesforce-build/references/choosing-the-tool.md, is relative to the skills folder that holds this skill.

Sources

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.