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

salesforce-delivery

Skill0 starsChanged 3 days agoMITView on GitHub

Run the work around a Salesforce build, phase by phase. Use it when you take over or assess an org, run discovery or write requirements and user stories, choose and record a solution design, set up or share sandboxes, plan testing or UAT, migrate data, plan a release or deployment, write documentation or training, hand over, support an org or review its health, or estimate and control scope. Gives for each phase what matters most, the artifact it produces and the human sign-off, with checklists, artifact contents, the costly failure modes and the practice that prevents each, and field lessons on verifying before stating, test side effects, other people's work, proportion and writing for clients.

The file

We split this skill into the rules it gives an agent

3 rules. About 0 are common, 1 is distinctive, and 2 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.

  • 2

    Very specific

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

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    After meaningful work: torque session add --workspace .

    line 74
  2. Very specific:

    When the consultant corrects you, or a fix took more than one attempt, save it as a lesson with its check (torque lesson --help), so the …

    line 76
  3. Distinctive:

    Not re-tested for this release.

    line 108

≈ “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 delivery

Find your phase in the map, do its few things that matter, produce its artifact, get its sign-off. State described: Winter '27 (API 68.0), checked 2026-11-14. Building itself is in skills salesforce-build and salesforce-flow.

Marks used in this skill: verify = not confirmed in a fetched primary source: verify in the org or the current documentation. field-reported = seen in real client work, usually in one org. [HUMAN] = a person decides or acts; the agent prepares.

The map

PhaseWhat matters mostProducesHuman sign-offRead
1. Take over or assess an orgResolve which org you are in. Inventory read-only, every figure dated. Change nothing while looking. Take a baselineOrg assessment, baseline record, access and environment registerClient admin fact-checks it; sponsor accepts scopereferences/assess-and-discover.md
2. Discovery and requirementsAsk why before how. Check every name against the org. Stories with pass or fail criteria. Non-functional needs and the success measure before designDiscovery record, stories with criteria, fit-gap, open-questions logStakeholders confirm the write-up; product owner approves the backlog and scopereferences/assess-and-discover.md
3. Solution designNative baseline and its effort first. Two options with trade-offs. One decision record per consequential choice. A design the client's admin can runSolution design, decision recordsA second person reviews; the client owner approves schema, access model and estimatereferences/design-and-estimate.md
4. EnvironmentsKnow each org's type and what a refresh destroys. In a shared org, retrieve, compare, deploy only yours. A sandbox can email real peopleEnvironment register, post-refresh runbookEach refresh; each merge to a protected branchreferences/environments-and-release.md
5. BuildThe two build skills. Enter every sandbox change in the build register the same dayBuild registerA second person reviews configuration and codereferences/environments-and-release.md
6. Testing and UATEach criterion: positive, negative, bulk, as each persona. What can a test send outside the org? UAT by people who do the job. Say what was not testedTest scripts, UAT log, UAT sign-offThe business owner signs UAT and decides go or no-goreferences/testing-and-uat.md
7. Data migrationPreparation is most of the job. Rehearse at full volume. Bypass list with restore steps. Reconcile with exact counts and control totalsMapping specification, load runbook, reconciliation reportThe data owner approves mapping, rehearsal and the production loadreferences/data-migration.md
8. ReleaseValidate, then deploy the validated job. Manual-steps register. Keep the current production version as the way back. Deployed is not active. Smoke-test as real usersDeployment runbook, manual-steps register, release notesThe client approver's go or no-goreferences/environments-and-release.md
9. Documentation, training, hand-overWrite for the reader's job. Dictionary from metadata. The admin guide says how to change things safely. Teach-backAdmin guide, user guides, data dictionary, decision log, hand-off noteThe client accepts the hand-over in writingreferences/handover-and-support.md
10. Support and org healthEvidence before cause. Fix the cause. Error mail goes to someone who reads it. A fix made by hand goes back to source. Review health each quarterTriage note, health review, debt registerThe reporter confirms the fix; the client owner agrees prioritiesreferences/handover-and-support.md
11. Estimating and scopeWritten assumptions. Count data preparation, testing, deployment and training. A new request is a change request. Compare effort with what the client was told before you polish a planEstimate with assumptions, change request logThe client signs the scope baseline; each change is decided before work startsreferences/design-and-estimate.md

Rules that hold in every phase

  1. State org facts from the org they are about, read that day. Label each with org and date. A sandbox is a dated copy. A negative ("nothing sends from here") is unverified until checked (field-reported: cards built from a sandbox were wrong about production).
  2. Say what "verified" rests on. Written, checked, reviewed and accepted are different facts. For a check, state as whom, in which org, by what kind of check, and what was not covered. A check counts only if its result would differ when the thing is wrong. A green deploy or an exit code is not the outcome, and acceptance is never the agent's to set (field-reported: 30% of work recorded as verified was contradicted later).
  3. Test on dummy records, never a real person's, and find out what a test can send outside the org before the first record.
  4. In anything other people use, add, do not overwrite. A sandbox, a flow, a sheet, a deck: take the current state, compare, change only what is yours.
  5. Proportion. The native way and its effort first. No stage, document, safeguard or review round that nobody asked for.
  6. Investigate when asked to investigate; carry through what you were asked to do. "Check whether we can fix it" gets findings and a proposal, not changes. A request to do work is carried through to a checked result, and its authorization holds until its scope changes or it is withdrawn. Ask at an unresolved consequential choice and before an outward or irreversible step, with the concrete result prepared.
  7. Client-facing text: vet once and post once. Restate the request first when its purpose or reader is unclear; outline to the reader first when the document is long or costly. A review loop that converges is not the reader accepting. A revision after review is a change list against the existing document.

Detail and the incidents behind these: references/working-practice.md. The 26 failure modes that cost consultancies most, each with its prevention: references/failure-modes.md.

Browser testing

To test as a user in a browser, sign in through the Salesforce front door with the CLI's existing authorization: torque browser login (see torque browser login --help). Do not type a password, do not ask the consultant to sign in to an org the CLI is authorized for, and never print or store a sign-in address. Then confirm whose session it is before you act (references/testing-and-uat.md, section 3).

Record as you go

  • At the start: torque context --workspace . --client NAME for the client's context and recent sessions, and the saved lessons (torque lesson --help). Then verify the cheap facts that drift: which org, which flow version is active.
  • After meaningful work: torque session add --workspace . --client NAME --summary TEXT --status prepared|executed|verified|incomplete. Keep job ids and evidence files.
  • When the consultant corrects you, or a fix took more than one attempt, save it as a lesson with its check (torque lesson --help), so the next session does not repeat it.
  • For the next consultant: torque handoff --workspace . --client NAME.
  • Find a recipe for a phase with torque workflows list --workspace ..

References

  • references/assess-and-discover.md: taking over an org, discovery, requirements.
  • references/design-and-estimate.md: options, decision records, estimates, scope control.
  • references/environments-and-release.md: sandboxes, shared orgs, the build register, release and deployment, deploy and retrieve mechanics.
  • references/testing-and-uat.md: test design, the checks that would have caught escaped defects, as-user testing, side effects, the UAT hand-over pack.
  • references/data-migration.md: mapping, rehearsal, loading, reconciliation.
  • references/handover-and-support.md: documentation, training, hand-over, support, org health, platform releases.
  • references/failure-modes.md: costly failure modes and what prevents each.
  • references/working-practice.md: what "verified" means, org facts, authority, other people's work, proportion, documents for clients.

A path that starts with another skill's name, such as salesforce-build/references/access-and-sharing.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.