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。
直接命令不会经过审查 Prompt；运行前请先检查来源。
npx skills add https://github.com/openshift/cluster-capi-operator --skill deep-review命令会保持在同一行。复制前请横向滚动并检查完整内容。
想先保存到本地？可下载 SkillsMP 当前能够提供的文件。
正在显示 SKILL.md
| name | deep-review |
| description | Run a two-pass parallel code review with synthesis |
| disable-model-invocation | false |
Run a two-pass code review of the changes described below.
Changes to review: $ARGUMENTS
Current branch: !git branch --show-current
Recent commits on this branch:
!git log --oneline -20
Before launching any agents, verify Gemini is available by running this smoke test using the Bash tool:
echo "Reply with exactly: OK" | gemini -p - 2>&1
If the command fails for any reason — binary not found, authentication error, configuration issue, network error, or unexpected output — Gemini is not available. In that case:
NOTE: Gemini is not available — the
geminiCLI is not working (not installed, not authenticated, or misconfigured). The Gemini review passes (agents 2 and 3) will be skipped. This review will use Claude-only analysis. Fix thegeminiCLI setup to enable multi-model review.
Before launching agents, assess the scope. If the changeset is large or spans distinct areas (e.g., framework code, platform-specific tests, migration logic), split the work across multiple agents by area rather than sending everything to one. Each agent should review a coherent subset.
Launch the following agents in parallel using the Agent tool, with run_in_background: true:
code-reviewer agent (subagent_type: code-reviewer) — Claude/Opus. Full code review:
gemini agent (subagent_type: gemini) — Gemini via CLI. Independent second opinion from a different model:
gemini agent (verifier) (subagent_type: gemini) — Second Gemini instance focused on verification:
If the changes include test files (_test.go, suite_test.go, or test helper files), ask the user whether to also launch a dedicated test quality reviewer. If yes, launch an additional agent:
subagent_type: code-reviewer) — test quality focus:
.claude/skills/test-standards/SKILL.md as the review checklistProvide all agents with sufficient context: commit range, relevant file paths, and what the feature does.
Wait for all agents to complete.
Once all agents have returned, synthesise their findings yourself (do NOT delegate this to a subagent). Analyse the sets of findings and:
Write the final review to .claude/reviews/review-$0.md, structured with sections that fit the findings. Use categories that match what was actually found rather than a fixed template. Common categories include:
Only include sections that have findings. Be specific — include file paths with line numbers and suggested diffs where applicable. Be concise — don't restate code that is already clear from the file path and line number.
Use when the user wants to audit, update, or regenerate RBAC permissions for a controller or service account. Covers: adding new permissions after code changes, auditing a service account for least privilege, periodic RBAC reviews, or regenerating RBAC manifests after adding new CRD types or controllers. Trigger on any mention of "rbac" in the context of updating or reviewing permissions.
Use when auditing, reviewing, or refining cloud provider CredentialsRequest permissions in OpenShift. Use when asked to minimize IAM permissions, check for unused permissions, or verify a CredentialsRequest manifest against provider source code. Triggers on mentions of CredentialsRequest, IAM permissions audit, credential refinement, or cloud provider permission minimization.
Apply when writing, modifying, or reviewing test files (_test.go). Covers test level selection, assertion quality, BDD naming, and use of shared helpers.
基于 SOC 职业分类
