second-opinion

name: second-opinion description: "Consolidated PingCAP review expert skill for local code review. Trigger when the user asks for a second opinion on a change (e.g., 'give second opinion on this change' or 'do code review for second opinion')."

Second Opinion

Asset Location (Claude Code only)

The review assets (taxonomy, experts, processes, policies, fragments, prompts) live in the skill's installation directory. Determine the base path by checking these locations in order and using the first one where taxonomy.md exists:

1. ~/.claude/skills/second-opinion/ (global install)
2. .claude/skills/second-opinion/ (project-local install)
3. The current working directory (when invoked inside the repo itself)

Read all referenced assets from that base path throughout this workflow.

Workflow

• Infer the review-output language from the user prompt that triggers second opinion.
• Use the inferred language for user-facing review communication.
• Do not ask a separate language-preference question unless the user asks to override the inferred language.
• Ensure all assets and code in this repo are in English, except for language-specific templates.
• If a subagent is available for the review stage, use it only when it provides context isolation; otherwise explicitly instruct it to ignore prior context and use only compiled_prompt + diff.
• Ignore sample foo assets under experts/foo, processes/foo, policies/foo.yaml, fragments/foo.md, and examples/foo; they are format references only.
• Run the review workflow (tagger → compiler → review) on the provided diff.
• The compiler deterministically selects processes and experts, and always includes policies.
• Write artifacts to the workspace root:
  • emit second_opinion_meta.json with selection rationale after the compiler stage
  • produce second_opinion.md and second_opinion.json after the review stage
• If GitHub posting is requested or confirmed, also produce second_opinion_github_comments.json in the workspace root.
• If posting was not requested initially, ask whether to post only after review outputs are complete.
• GitHub comment posting must use English, prefer inline/file comments over general comments, and end comments with Source: <type>/<id> | [Second Opinion](https://github.com/PingCAP-QE/second-opinion).
• File comments must use PR review comment semantics (subject_type=file), and inline failures due to unresolved lines should retry as file comments.
• Do not duplicate one finding across multiple comment modes, and do not add an extra review body comment per finding when inline/file comments are posted.
• For actual posting, use gh CLI; if gh is unavailable or unauthenticated, fail posting explicitly.
• Chat findings are supplementary; they do not replace required file outputs.