Loading

Agentforce: Common Agentforce Behavior Issues and How to Resolve Them

Veröffentlichungsdatum: Jul 1, 2026
Beschreibung

Agentforce behavior issues occur when an agent is active and accessible, but its responses or execution do not match expectations.

Common scenarios include:

  • The agent does not invoke the expected Subagent(Topic) or Action
  • Responses are incomplete, truncated, or rewritten
  • Knowledge citations do not appear in responses
  • The agent escalates unexpectedly to a human
  • Identical inputs produce inconsistent outputs

These issues are usually caused by configuration or setup rather than system errors.

This article provides a self-service troubleshooting guide to help you identify and resolve common Agentforce behavior issues using Agent Builder, configuration settings, and setup validation


How to Troubleshoot Agentforce Issues

When Agentforce behavior is unexpected, always follow this order:

Step 1: Subagent Selection

  • Was the correct Subagent selected?
  • If not, review Subagent naming, scope, and overlap.

Step 2: Action Execution

  • Was the expected Action invoked?
  • If not, check Action descriptions and usage instructions.

Step 3: Data and Knowledge Output

  • Is the response missing or incomplete data?
  • Verify Knowledge and Action output configuration.

Step 4: Instructions and Prompting

  • Are instructions conflicting or too complex?
  • Simplify or clarify Subagent instructions.

Step 5: System Constraints

  • Check for token limits, execution limits, or streaming issues.

Root Causes of Agentforce Behavior Issues

Most issues fall into one of the following categories:

1. Routing Issues (Subagent Selection)

  • Overlapping Subagents
  • Generic or unclear Subagent names
  • Broad classification descriptions

2. Action Selection Issues

  • Similar or duplicate Action definitions
  • Missing “when to use” guidance
  • No explicit Action mapping in instructions

3. Prompting and Instruction Issues

  • Conflicting instructions
  • Overly complex or lengthy prompts
  • Business logic embedded in text instead of configuration

4. Data and Grounding Issues

  • Knowledge articles not published or indexed
  • Citations not enabled
  • Incomplete Action outputs

5. System Constraints

  • Token limits
  • LLM execution limits
  • Streaming or rendering issues
Lösung

Prompting Issues

Prompting issues are the most frequent cause of unexpected Agentforce behavior. Unlike deterministic systems such as traditional Einstein Bots, LLM-based agents can behave unpredictably if prompts are unclear or inconsistent.

Where Prompting Occurs

Agent behavior is influenced by prompts defined in:

  • Subagent descriptions

  • Subagent instructions

  • Action descriptions

  • Action input descriptions

  • Action output descriptions

Best Practices

  • Use explicit and unambiguous language in Subagent and Action descriptions

  • Avoid relying on the LLM to infer business rules

  • Clearly define when an Action should and should not be used

  • Ensure Action outputs are clearly structured and described


Session Won’t Start or “Something went wrong. Refresh the conversation and try again”

If conversations fail to start or display generic errors, validate the following setup items.

Troubleshooting Steps

Toggle Agentforce

  • Navigate to Setup → Einstein Copilot

  • Turn Agentforce off, save, then turn it back on

Toggle Einstein Bots

  • Navigate to Setup → Einstein Bots

  • Turn off, save, then turn on

Toggle Einstein Setup

  • Navigate to Setup → Einstein GPT Setup

  • Turn off, save, then turn on

Verify Default Agent Exists

  • Ensure “Enable the Agentforce (Default) Agent” is enabled

  • Toggle it off and on again if needed


Topic or Action Not Getting Called


Common Causes

  • Prompting issues

  • Topic or Action filters

  • Missing context variables

  • Permission restrictions

What to Check

Topic and Action Filters

  • Confirm filters are correctly configured

  • Ensure required context variables exist before evaluation

Important Considerations

Context variables must never be assumed or derived from prompt instructions.

They are not automatically available in Agent Builder and must always be explicitly set using deterministic mechanisms such as:

  • Action output mapping

  • Variable mapping in configuration

  • Flow or structured assignment logic

Do not rely on prompt instructions, Subagent instructions, or natural language reasoning to set or infer context variables.

Any logic that depends on context variables must ensure they are explicitly initialized before evaluation.


Citations Not Appearing in Responses

Citations are displayed only when:

  • The Knowledge Action returns citations

  • Citations are enabled for that Action

Troubleshooting Steps

  • Open Knowledge Action configuration

  • Confirm citations are enabled

  • Test the Action independently

  • Ensure response is not overwritten later in the flow

Also verify:

  • Knowledge articles are published

  • Articles are indexed

  • Articles are accessible to the agent


Response or Content Is Getting Truncated


Limits

AreaLimit
LLM response~2048 tokens
Action output~65,000 characters

 

Symptoms

  • Responses cut off mid-sentence

  • Missing sections in long outputs

  • Partial Action results

Mitigation

  • Use “Show in Conversation” instead of embedding large outputs

  • Avoid returning unnecessary fields in Actions

  • Break complex requests into multiple steps

  • Store large data in objects and reference via Actions


Streaming Issues


Symptoms

  • “Something went wrong” messages

  • Responses appear then get rewritten

  • Partial or inconsistent outputs

Troubleshooting Steps

  • Review prompt templates for strict formatting (for example, JSON-heavy outputs)

  • Simplify complex or nested instructions

  • Reduce large Action outputs

  • Consider switching models if instability persists

Best Practices

  • Avoid generating large responses in a single turn

  • Break workflows into smaller steps

  • Define fallback responses to prevent repeated regeneration


Agent Response Gets Rewritten

If responses appear and then change or disappear, groundedness checks may be triggered.

Mitigation

  • Ensure instructions are accurate and specific

  • Avoid overly broad or ambiguous prompts

  • Ensure responses are grounded in Knowledge or structured data


Subagent(Topic) or Action Not Available


Common Reasons

  • Filtering logic excludes the Topic or Action

  • Permission restrictions

  • Missing configuration in agent setup

Troubleshooting Steps

  • Review filter logic carefully

  • Ensure required context variables are set

  • Confirm Topic or Action is included in agent configuration

  • Validate permissions with administrator if needed


URL Redaction in Responses

Some URLs may be hidden for security reasons.

Resolution

  • Add trusted URLs in Setup → Trusted URLs

  • Ensure URLs are returned by Actions or explicitly allowed


Unexpected Human Escalation

Escalation may occur due to:

  • Agent unable to proceed

  • Escalation Topic selection

  • System limits reached

Recommendations

  • Review escalation logic and naming

  • Ensure fallback Topics exist

  • Validate escalation conditions are intentional


Agent Exceeds Maximum LLM Calls

An agent can make up to 8 LLM calls per user turn.

Common Causes

  • Repeated Action failures

  • Validation loops

  • Overly complex workflows

Mitigation

  • Simplify Subagent flows

  • Reduce retry logic

  • Break workflows into multiple steps


Flow Execution Errors

Error Message:
“An error occurred when executing a flow interview”

Resolution

Ensure the Service Agent User has:

  • Agentforce Service Agent Secure Base

  • Agentforce_Service_Agent permissions

  • Permission Set Group: AgentforceServiceAgentUserPsg

  • Permission Set License: Agentforce Service Agent User


Language or Locale Issues

If the agent responds in the wrong language:

  • Verify agent language configuration

  • Check org locale settings

  • Ensure prompts are language-consistent


Global Instructions

Global Instructions define system-level behavior for Agentforce.

Recommended Uses

  • Maintain consistent tone and style

  • Define fallback behavior

  • Handle ambiguity

  • Manage sensitive scenarios

  • Address competitor mentions


Most Agentforce behavior issues are caused by configuration issues such as:

  • Overlapping Subagents

  • Unclear Action definitions

  • Conflicting instructions

  • Missing or unindexed Knowledge

Improving Subagent boundaries, Action clarity, and instruction design resolves the majority of issues

Nummer des Knowledge-Artikels

005305511

 
Laden
Salesforce Help | Article