Answer: To write an eval case manually, run `claude plugin eval init --bare first-case` to create a template, then write the prompt and run limits in `prompt.md` frontmatter (e.g., `max_turns` and `allowed_tools`) and add graders under `graders/` such as an `llm` criteria.md or a `tool_used` skill-fired.md. Run the case with `claude plugin eval .` from the plugin root. Status: Confirmed Entry: tool/claude-code/agents-and-automation/write-eval-case Revision: 3 Digest: sha256:c53103b269fef785 (changes whenever the content changes) Timeline: Created: 2026-11-13T03:04:54Z Last changed: 2026-11-19T12:18:11Z (skill-fired grader type: — → type: tool_used) Last verified: 2026-11-20T12:22:41Z Next check: 2026-11-21T12:22:41Z Check interval: Daily Expected change: No fixed schedule Data: Field | Value | Source | As of bare init command | claude plugin eval init --bare first-case | [1] | 2026-11-20 prompt.md frontmatter example | max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] | [1] | 2026-11-20 grader frontmatter type | type: llm | [1] | 2026-11-20 skill-fired grader type | type: tool_used | [1] | 2026-11-20 run command | claude plugin eval . | [1] | 2026-11-20 Sources: [1] Anthropic. "Test plugins with evals". https://code.claude.com/docs/en/plugin-evals.md Role: primary. Accessed: 2026-11-20T12:22:41Z Quote: "claude plugin eval init --bare first-case" (Supports: bare init command) Quote: "max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill]" (Supports: prompt.md frontmatter example) Quote: "type: llm" (Supports: grader frontmatter type) Quote: "type: tool_used" (Supports: skill-fired grader type) Quote: "with `claude plugin eval .` from the plugin root" (Supports: run command) Published: 2026-11-20T03:14:09Z Snapshot SHA-256: f612b7a35aa589f2beeeded25d8143492fde96a41688bf016856679a58211a98 (captured 2026-11-19T12:18:12Z) Verification: Method: Automated extraction from [1] with deepseek-flash (JSON output, thinking disabled). Every value was checked by code to appear verbatim in a quoted passage of [1]. On this check the text of [1] was unchanged since 2026-11-19T12:18:11Z, so the values verified then were kept. Cross-check: not-available at 2026-11-20T12:22:41Z. No independent second source is configured for this entry Last successful check: 2026-11-20T12:22:41Z Last attempt: 2026-11-20T12:22:41Z Notes and limitations: Extracted automatically from a single official source (section "Write a case manually"). Related: https://latestn.com/tool/claude-code/agents-and-automation/write-mod-test How do I write an automated test for a Claude Code mod? https://latestn.com/tool/claude-code/agents-and-automation/zod-pydantic-schemas How do I define a type-safe schema with Zod or Pydantic for Claude Code structured outputs? https://latestn.com/tool/claude-code/commands-and-shortcuts/available-actions-contexts Which keyboard shortcut contexts and actions are available in Claude Code? https://latestn.com/tool/claude-code/agents-and-automation/what-is-a-hook What is a hook in Claude Code? https://latestn.com/tool/claude-code/agents-and-automation/what-is-a-subagent What is a subagent in Claude Code? https://latestn.com/tool/claude-code/agents-and-automation/when-to-use-single-message-input When should I use single message input instead of streaming input in the Claude Agent SDK? https://latestn.com/tool/ All entries in AI tools Change log: Time | Event | Change | Source 2026-11-19T12:18:11Z | changed | skill-fired grader type: — → type: tool_used | [1] 2026-11-19T12:18:11Z | changed | grader frontmatter type: — → type: llm | [1] 2026-11-19T12:18:11Z | changed | prompt.md frontmatter example: — → max_turns: 10 allowed_tools: [Read, Glob, Grep, Skill] | [1] 2026-11-19T12:18:11Z | changed | grader type (tool_used): type: tool_used → — | [1] 2026-11-19T12:18:11Z | changed | grader type (llm): type: llm → — | [1] 2026-11-19T12:18:11Z | changed | allowed_tools field: allowed_tools: [Read, Glob, Grep, Skill] → — | [1] 2026-11-19T12:18:11Z | changed | max_turns field: max_turns: 10 → — | [1] 2026-11-18T12:18:01Z | changed | grader type (tool_used): — → type: tool_used | [1] 2026-11-18T12:18:01Z | changed | grader type (llm): — → type: llm | [1] 2026-11-18T12:18:01Z | changed | skill-fired grader type: type: tool_used → — | [1] 2026-11-18T12:18:01Z | changed | grader type: type: llm → — | [1] 2026-11-13T03:04:54Z | added | Entry created | — Changes for this entry as a feed: https://latestn.com/tool/claude-code/agents-and-automation/write-eval-case/latest Challenge this entry: Open challenges: None How: If a value here is wrong or out of date, send what is wrong and a link to evidence. Challenges are queued and checked against the evidence; they do not change the entry by themselves. If the entry changes as a result, the change log says so. Form: https://latestn.com/dispute/tool/claude-code/agents-and-automation/write-eval-case POST: https://latestn.com/dispute/tool/claude-code/agents-and-automation/write-eval-case with claim, evidence_url, and optionally field and contact (form-encoded or JSON) GET: https://latestn.com/dispute/tool/claude-code/agents-and-automation/write-eval-case?claim={what is wrong}&evidence_url={link} How to cite: LatestN. (2026). How do I write an eval case for a Claude Code plugin, including prompt.md frontmatter and case.yaml fields? Revision 3, last verified 2026-11-20T12:22:41Z. https://latestn.com/tool/claude-code/agents-and-automation/write-eval-case (accessed 2026-11-21T10:00:10Z). @misc{LatestN_tool_claude_code_agents_and_automation_write_eval_case_r3, title = {How do I write an eval case for a Claude Code plugin, including prompt.md frontmatter and case.yaml fields?}, author = {{LatestN}}, year = {2026}, howpublished = {\url{https://latestn.com/tool/claude-code/agents-and-automation/write-eval-case}}, note = {Revision 3; digest sha256:c53103b269fef785; last verified 2026-11-20T12:22:41Z; accessed 2026-11-21T10:00:10Z} } Reuse: License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) Attribution: Copying and adapting are allowed, including commercially, if you credit LatestN and link to the entry you used. Each entry ends with a ready-made citation. URL: https://latestn.com/tool/claude-code/agents-and-automation/write-eval-case