sf-ai-agentforce-testing_skill

Overview

This skill is a comprehensive Agentforce testing toolkit that runs a dual-track workflow: multi-turn API testing (primary) and CLI Testing Center runs (secondary). It executes multi-turn conversations, generates YAML test specs and scenarios, analyzes coverage across topics/actions/context, and performs automated fix-and-retest cycles with a 100-point, 7-category scoring model. The skill is implemented in Python and integrates with other skills for fixes, test data, and observability.

How this skill works

The skill discovers agents, retrieves agent metadata, and either runs multi-turn sessions against the Agent Runtime API or executes single-utterance tests via the sf CLI. It generates categorized multi-turn scenarios and test specs, executes tests (sequential or swarm), parses results, scores across seven validation categories, and can invoke an automated agentic fix loop to remediate failures and re-run tests. Credential and environment resolution is automated for API testing; CLI tests require the Agent Testing Center feature in the org.

When to use it

• Validate multi-turn conversation flows, context preservation, and topic re-matching.
• Run quick single-utterance smoke tests via the sf CLI when Testing Center is enabled.
• Integrate automated tests into CI/CD pipelines using the Python scripts.
• Perform coverage analysis across topics, actions, and context variables.
• Execute automated test-fix loops to improve agent quality before production release.

Best practices

• Run dependency deployment (Apex, Flows) and seed test data before testing to avoid false failures.
• Use multi-turn API testing as the primary track for conversation, context, and re-matching checks.
• Keep a test-plan-{agent}.yaml to skip the interview flow and reproduce test suites consistently.
• Delegate agent script fixes to the designated agent-script repair workflow and test data to the data skill.
• Start with CLI tests for fast smoke validation, then run multi-turn API tests for deep coverage.

Example use cases

• Run a full multi-turn test suite to verify topic switching and escalation paths for a contact center agent.
• Execute CLI-based single-utterance tests for nightly smoke checks in a CI pipeline.
• Generate multi-turn scenarios from agent metadata to validate guardrails and safety instructions.
• Use the automated fix loop to iteratively resolve failing tests and improve the agent score.
• Perform coverage reporting to identify topics or actions missing test coverage before release.