AgentSkillsCN

generate-tests

生成 EvalView 测试用例——可通过 LLM 驱动的生成从 SKILL.md 文件中生成，或通过代理捕获真实代理交互。

SKILL.md
--- frontmatter
name: generate-tests
description: Generate EvalView test cases — either from a SKILL.md file using LLM-powered generation, or by capturing real agent interactions through a proxy.

Generate Tests

Use this skill when the user wants to create test cases for their AI agent or skill without writing YAML by hand.

Four approaches

1. Generate tests from a SKILL.md file

Use the generate_skill_tests MCP tool to auto-generate a test suite from a skill definition. This reads the SKILL.md and produces YAML test cases covering explicit triggers, implicit triggers, contextual triggers, and negative cases.

Steps:

  1. •Ask the user which SKILL.md to generate tests for (or detect it from context).
  2. •Call generate_skill_tests with:
    • •skill_path: path to the SKILL.md file
    • •output_path (optional): where to save the generated YAML
    • •count (optional): number of test cases (default: 10)
  3. •After generation, offer to run the tests with run_skill_test.

CLI equivalent:

code
evalview skill generate-tests .claude/skills/my-skill/SKILL.md --auto
evalview skill generate-tests .claude/skills/my-skill/SKILL.md -c 20 -o tests/my-skill-tests.yaml

2. Create individual test cases manually

Use the create_test MCP tool to create a single test YAML file from a description.

Steps:

  1. •Gather from the user: test name, query, expected tools, forbidden tools, expected output keywords, and minimum score.
  2. •Call create_test with the parameters.
  3. •After creating the test, call run_snapshot to establish the golden baseline.

3. Capture real interactions

Use the CLI evalview capture command to proxy real agent traffic and save interactions as test YAMLs automatically. This records the query, output, and tool calls from live usage.

CLI equivalent:

code
evalview capture --agent http://localhost:8080/execute --output-dir tests/test-cases
evalview capture --multi-turn  # saves all turns as one multi-turn conversation test

4. Validate a skill before testing

Use validate_skill to check a SKILL.md for correct structure and completeness before generating tests from it.

Running generated tests

After generating tests, execute them with run_skill_test:

  • •test_file: path to the generated YAML
  • •no_rubric: true for fast deterministic-only checks (no LLM cost)
  • •verbose: true for detailed output on all tests

CLI equivalent:

code
evalview skill test tests/my-skill-tests.yaml
evalview skill test tests/my-skill-tests.yaml --no-rubric  # fast, $0
evalview skill test tests/my-skill-tests.yaml --verbose --model claude-sonnet-4-20250514