一键导入
review
Use when the user asks to review the current branch, check code quality, run a code review, or assess changes before merging.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
菜单
Use when the user asks to review the current branch, check code quality, run a code review, or assess changes before merging.
用 Codex 或 Claude 帮你安装 复制这段 Prompt,粘贴到 Codex、Claude 或其他助手里,让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
Use when working inside an api-cp-* repo and the user wants to publish/expose their API docs, "add my API to the catalog", set up the publish-api-docs workflow, or get a hmcts.github.io docs site. Wires the current repo into the amp-catalog Swagger UI pipeline, handles the two one-time GitHub Pages admin gaps that fail silently otherwise, and verifies the live site. This is the per-repo counterpart to the catalog-publisher agent, which only touches amp-catalog/docs/apis.json.
Wire up auto-dev and auto-SIT deployment CI for a service-cp-* repo after Azure provisioning and cp-vp-aks-deploy registration are complete. Idempotent — safe to invoke again if the jobs already exist. Use when a service-cp-* repo is missing the deploy-dev and deploy-sit jobs in its ci-build-publish.yml, or when setting up a newly bootstrapped service for the first time.
Start a new HMCTS API-Marketplace API specification repo using the canonical HMCTS template (api-hmcts-crime-template) as master source. Use when creating a new api-cp-* repo (OpenAPI-first, spec-only) distinct from a runtime service-cp-* service.
Stand up a new HMCTS API-Marketplace service-cp-* using the canonical HMCTS template (service-hmcts-crime-springboot-template) as master source. Use when creating a new Spring Boot service for the Common Platform.
Bootstrap the gitignored .claude/CLAUDE.md for any api-cp-* or service-cp-* repo with three @import lines pointing to hmcts-apim-sdlc-orchestrator context standards. Runs automatically on every session start via the SessionStart hook — invoke manually only to force an update.
Use when the user wants to review, audit, or check an OpenAPI v3 spec — applies HMCTS data-sharing policy (UK GDPR/DPA 2018), infrastructure SLA (Azure APIM/AKS), API standards, and security standards (OAuth 2, TLS, input validation). Returns a scored report with Critical/Warning/Info findings.
| name | review |
| description | Use when the user asks to review the current branch, check code quality, run a code review, or assess changes before merging. |
Run a comprehensive review of the current branch by orchestrating three review passes in parallel, then aggregating their findings.
Run the following to check which plugins are available:
claude plugin list --json 2>/dev/null | python3 -c "
import json,sys
installed = {p['id'] for p in json.loads(sys.stdin.read()).get('installed', [])}
for pid in ['pr-review-toolkit@claude-plugins-official','code-review@claude-plugins-official']:
print('ok' if pid in installed else 'missing', pid)
"
For any plugin reported as missing, tell the user:
To get the full 3-reviewer report, install the missing plugin(s) and reload:
/plugin install <name>@claude-plugins-official /reload-pluginsContinuing with the reviewer(s) that are available.
Do not block the review — proceed with whichever reviewers are present. The diff review always runs regardless.
Invoke all three of the following simultaneously using the Agent tool (one agent per review), passing the same PR or branch context to each:
Diff review — run git diff main...HEAD, then for each changed file assess:
PR Review Toolkit — invoke the pr-review-toolkit:review-pr skill using the Skill tool. Skip if not installed (noted in Step 0).
Official code-review — invoke the code-review:code-review skill using the Skill tool. Skip if not installed (noted in Step 0).
Collect results from all three passes and deduplicate. Present a single unified report:
Blockers that should prevent merge (any finding rated critical/blocker across any reviewer).
Non-blocking improvements worth addressing before merge.
Style or minor quality observations (low priority).
What the reviewers agreed was well done.
If there are no issues in a category, state "None found."
End with a single overall recommendation: Approve, Request changes, or Needs discussion.
Note which reviewers agreed and which (if any) diverged.
Keep the final report concise and actionable. Reference specific file paths and line numbers.