Agentforce Testing Checklist
This Agentforce testing checklist gives Salesforce teams an itemised, layer-by-layer list of what to verify before an AI agent goes live and after every change. It turns the twelve test layers of a complete agent program into concrete checks you can work through and sign off.
An active-roadmap capability, available today for demonstration and proof-of-concept validation with Salesforce testing teams.
Why a checklist, and how to use it
An Agentforce agent is not a single feature to test; it is a stack of behaviours, from understanding a request to completing a business outcome across systems. Without a structured list, teams tend to test what is easy to see, the happy-path conversation, and miss the layers that cause the worst production incidents: grounding, security and handoff. A checklist makes coverage explicit so nothing is skipped by accident.
Use this checklist twice. First as a go-live gate, working every layer before an agent reaches production. Then as a change gate, re-running the affected layers whenever a prompt, topic, action, model version or seasonal Salesforce release such as Spring, Summer or Winter changes agent behaviour. Each layer here has a dedicated deep-dive page; this checklist is the map that ties them together. For the full framework and rationale, see the Agentforce testing pillar.
A complete run should produce evidence, not just a set of green checks. Record who tested each layer, against which agent version and data, and what the result was, so sign-off is defensible and the next release starts from a known baseline.
The twelve-layer Agentforce testing checklist
Work top to bottom. Each layer lists what to verify and links to its deep-dive where one exists.
| # | Test layer | What to verify | Status |
|---|---|---|---|
| 1 | Intent recognition | The agent understands what the user is actually asking across informal, ambiguous and multi-part requests. | □ |
| 2 | Topic routing | Requests reach the correct topic with the right instructions and permitted actions. See topic testing. | □ |
| 3 | Utterance consistency | Varied phrasings of the same request behave equivalently. See utterance testing. | □ |
| 4 | Action selection | The agent chooses the right action and avoids running ones it should not. See action testing. | □ |
| 5 | Flow & Apex execution | Invoked actions produce the correct system result end to end. See Flow & Apex action testing. | □ |
| 6 | Grounding & data accuracy | Responses are backed by approved data, not invented. See grounding testing. | □ |
| 7 | Response quality | Answers are accurate, complete and appropriate for the request and audience. | □ |
| 8 | Security & access | Permissions, sensitive-data exposure, prompt injection and tool misuse are covered. See security testing. | □ |
| 9 | Human handoff | Escalation happens at the right moment with full context. See human handoff testing. | □ |
| 10 | Business-outcome completion | The end-to-end task finishes correctly across every system involved, not just a plausible reply. | □ |
| 11 | Performance & resilience | Acceptable latency under realistic load and graceful recovery when an action or lookup fails. | □ |
| 12 | Regression & monitoring | Suites re-run on every change and production is monitored. See regression testing. | □ |
Want this as a working document?
A structured version of this checklist, with per-item acceptance criteria and an evidence column for sign-off, is available on request. Ask for it when you schedule a Salesforce testing demo and we will walk through it against your agent.
Request the ChecklistTurning the checklist into repeatable coverage
SyntraFlow is designed to operationalise every line of this checklist as versioned, re-runnable tests, and to complement Salesforce-native tooling rather than replace it.
One suite per layer
Each checklist layer maps to its own test suite, so coverage gaps are visible instead of implied.
Go-live and change gates
The same checklist runs as a release gate before launch and a change gate on every prompt, action, model or seasonal release update.
Evidence for sign-off
Results are designed to be captured as audit evidence, recording tester, version and outcome for defensible sign-off.
Native-tooling friendly
Built to work alongside Agentforce Testing Center, the Testing API and Agentforce DX, orchestrating their runs into repeatable suites.
Cross-application coverage
Outcome and action layers extend past Salesforce to confirm results in connected systems.
A living baseline
Each completed run becomes the baseline the next release is measured against, so quality is tracked over time.
The checklist does not end at Salesforce
Layers 5 and 10, action execution and business-outcome completion, frequently reach into Oracle, SAP, NetSuite or Workday. A checklist that stops at the Salesforce record can be fully green while the transaction in the system of record is wrong or missing.
SyntraFlow is designed to carry the outcome-completion checks across application boundaries, pairing Agentforce testing with Oracle ERP testing so end-to-end sign-off reflects the real result everywhere it lands.
What a checklist-driven program gives you
Qualitative outcomes teams can expect when Agentforce testing follows a structured checklist.
Complete coverage
Every layer is accounted for, so the high-risk ones are never skipped in favour of easy happy paths.
Defensible sign-off
Recorded evidence per layer makes go-live and change approvals auditable rather than anecdotal.
Faster, safer change
Knowing which layers a change affects lets teams re-run the right subset and ship with confidence.
A shared language
The twelve layers give admins, developers and testers one vocabulary for what "tested" means.
A tracked baseline
Each run sets the reference point for the next, turning quality into a trend instead of a snapshot.
Cross-system assurance
Outcome checks reach into connected ERPs, so sign-off reflects the real end-to-end result.
Agentforce testing checklist FAQs
What does the Agentforce testing checklist cover?
It covers twelve layers: intent recognition, topic routing, utterance consistency, action selection, Flow and Apex execution, grounding, response quality, security and access, human handoff, business-outcome completion, performance and resilience, and regression and monitoring. Each layer lists what to verify so coverage is explicit.
How do I get a copy of the checklist?
A structured version with per-item acceptance criteria and an evidence column for sign-off is available on request. Ask for it when you schedule a Salesforce testing demo, and the team will walk through it against your own agent rather than sending a generic document.
When should the checklist be run?
Run it as a go-live gate before an agent reaches production, then as a change gate whenever a prompt, topic, action, model version or seasonal Salesforce release such as Spring, Summer or Winter alters behaviour. Re-run at least the layers a given change can affect.
Do I need to test all twelve layers every time?
Before go-live, yes. For an incremental change, re-run the layers the change touches plus a regression pass on the outcome and security layers, since those catch the most damaging regressions. The checklist helps you scope the right subset rather than guessing.
Does the checklist replace Agentforce Testing Center?
No. Agentforce Testing Center, the Testing API and Agentforce DX are Salesforce-native tooling for defining and running agent tests. This checklist organises what to test across all twelve layers, and SyntraFlow is designed to operationalise it alongside those native tools, adding cross-application coverage and audit evidence.
How does the checklist handle cross-system outcomes?
The action-execution and business-outcome layers include verifying results in connected systems such as Oracle, SAP, NetSuite or Workday. SyntraFlow is designed to carry those checks across application boundaries so a green checklist reflects the real end-to-end result, not just the Salesforce record.
Get the full checklist and a working demo
Schedule a Salesforce testing demo and we will share the structured, sign-off-ready checklist and walk every layer against your own Agentforce agent.