[karanb192]

Summary

The tool_called examples in the writing guide use server::tool names (github::create_pull_request, project-mgmt::create_publish_plan_simple). tool_called compares names exactly, and Claude Code, Codex, and Qoder CLI record MCP tools as mcp__<server>__<tool>, so a rule copied from these examples never matches. The skill-upper judge reference has the same example, so agents that write evals with skill-upper copy it too.

This PR switches the examples to the recorded form and adds a short list of how a tool_responses mock behaves.

Related issues

None. Docs-only fix, so I went straight to a PR as CONTRIBUTING allows.

Changes

• docs/guide/writing-evals.md and docs/zh/guide/writing-evals.md:
  • tool_called examples use mcp__github__create_pull_request and mcp__project-mgmt__create_publish_plan_simple.
  • New note after the rule_based evaluation order: tool names match exactly, in the mcp__<server>__<tool> form.
  • New list under MCP configuration: stdio only; started with node, which must be on PATH (skill-up sets up Node.js only when it installs a Node-based engine CLI itself); one default response per tool; generic tool description and open input schema; an error for undeclared tools.
• skills/skill-upper/references/judge-types.md: the same example fix plus a one-line note.

Test plan

• make test passes
• make verify passes (fmt + vet + lint)
• npm run docs:build passes
• Name forms checked against the code: codexMCPToolName in internal/agent/codex.go, the Qoder stream fixture in internal/agent/qodercli_test.go, and e2e/testdata/sandbox-mcp, which already uses mcp__agent-sandbox__.... The Claude Code adapter passes the stream name through, and Claude Code names MCP tools mcp__<server>__<tool> (Claude Code MCP docs).

Notes for reviewers

• I left internal/judge/rule_based_test.go alone. It uses github::... names on synthetic transcripts and only tests exact matching.
• An inline {{params...}} breaks the VitePress build, so the template bullet points to the example's name field instead.
• I did not state the name form for Qwen Code or OpenCode because I could not confirm it from the tests.
• The Chinese text is my translation. Please correct the wording if needed.

[chatgpt-codex-connector]

Codex Review Summary

This comment shows the latest Codex review activity on this pull request.

Review	Status	Commit	Review trigger
📝 Code Review	✅ Completed 2026-10-25T16:44:38.327836Z	fb553b5	PR opened

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review" or "@codex security review".

Codex reacts with 👀 while any review is running, comments if it has suggestions, and reacts with 👍 once all reviews finish with no findings.

[CLAassistant]

Thank you for your submission! We really appreciate it. Like many open source projects, we ask that you sign our Contributor License Agreement before we can accept your contribution.

---

Karan Bansal seems not to be a GitHub user. You need a GitHub account to be able to sign the CLA. If you have already a GitHub account, please add the email address used for this commit to your account.
_{You have signed the CLA already but the status is still pending? Let us recheck it.}

[zpzjzj]

LGTM