API审查
- 作者仓库星标 0
- 作者更新于 实时读取
- 作者仓库 skills-registry
- 领域
- 安全
- 兼容 Agent
-
- Claude Code
- Cursor
- Cline
- Codex
- Windsurf
- Gemini CLI
- +20
- 信任分
- 88 / 100 · 社区维护
- 作者 / 版本 / 许可
- @tomevault-io · 未声明 license
- Token 消耗评级
- 低消耗
- 接入复杂程度
- 即装即用
- 是否需要外部 API Key
- 不需要
- 兼容的系统
- macOS · Linux · Windows
- 底层运行要求
- 无特殊要求
- 文件与系统权限
-
- 只读
- 允许写入 / 修改
- 网络行为
- 允许外网请求
- 安装命令数
- 26 条
档案由构建时根据 SKILL.md 与安装命令自动衍生,可能与作者实际意图存在差异。
需要注意: 未限定 allowed-tools,默认拥有全部工具权限。
---
name: code-review
description: Review code changes, diffs, commits, branches, or PRs for correctness, regressions, security-sen…
category: 安全
runtime: 无特殊运行时
---
# code-review 输出预览
## PART A: 任务判断
- 适用问题:安全审计、密钥扫描、权限检查或风险分析。
- 输入要求:目标材料、限制条件、期望输出和验收方式。
- 证据边界:围绕“Contract / Pick Review Target / Dirty local changes”读取原文规则,不把推断写成作者承诺。
## PART B: 执行结果
- **01** 任务判断:确认你的需求是否属于安全审计、密钥扫描、权限检查或风险分析,并标出输入、限制和预期结果。
- **02** 执行计划:优先按“Contract / Pick Review Target / Dirty local changes”拆成步骤,说明每一步会读取什么、修改什么、产出什么。
- **03** 交付结果:给出可复制的命令、文件改动、检查清单或内容草稿,并说明如何继续迭代。
- **04** 风险边界:结合 读取文件、写入/修改文件、会按任务需要访问外部网络、通常不需要额外 API Key 给出执行前确认项。
## Running Rules
- 读取文件、写入/修改文件;会按任务需要访问外部网络;通常不需要额外 API Key。
- 先小样例验证,再放大到真实任务。
- 交付时同时给结果、检查口径和下一步迭代建议。 原文没有稳定的斜杠命令要求。安装验证后通常全局生效,直接在对话里点名这个 Skill 并描述任务即可。
告诉 Agent 目标文件或材料、期望结果、不可改范围、是否允许联网或执行命令。本 Skill 的权限画像是:读取文件、写入/修改文件。
先用一个小任务确认它会围绕“Contract / Pick Review Target / Dirty local changes”工作;涉及文件或命令时,先看 diff、日志、预览或测试结果。
检查最终产物是否包含明确结果、必要证据和下一步动作;如果输出泛泛而谈,就补充输入、边界和验收标准后重跑。
---
name: code-review
description: Review code changes, diffs, commits, branches, or PRs for correctness, regressions, security-sen…
category: 安全
source: tomevault-io/skills-registry
---
# code-review
## 什么时候使用
- 用于审阅代码、文档或方案并给出可执行反馈 适合处理安全审计、密钥扫描、权限检查和风险分析,核心价值是把输入、判断、执行、验证和交付边界固定下来,避免 Agent 泛泛回答。 把任务拆成可执行、可检查、可继续迭代的步骤;通常不需要额外 A…
- 面向安全审计、密钥扫描、权限检查或风险分析,优先处理能明确输入、步骤和验收标准的工作。
## 需要提供什么
- 目标材料、目录范围、期望结果和不可改动内容。
- 是否允许联网、执行命令、读写文件或调用外部服务。
## 执行规则
- 围绕「Contract / Pick Review Target / Dirty local changes」组织步骤,不把推断写成作者事实。
- 读取文件、写入/修改文件;会按任务需要访问外部网络;通常不需要额外 API Key。
- 先跑小样例,确认结果可检查后再扩大任务范围。
## 输出要求
- 给出最终产物、关键证据、验证方式和下一步动作。
- 信息不足时标记 unknown,不编造命令、平台或依赖。 作者原文负责流程事实;仓库文件负责来源和命令;流狐只补充适用场景、限制和质量判断。
skill "code-review" {
输入层 -> 用户目标 + 目标文件 + 禁止范围 + 验收标准
上下文层 -> Contract / Pick Review Target / Dirty local changes
规则层 -> SKILL.md 触发条件 / 执行顺序 / 输出格式
运行层 -> 无特殊运行时 | 读取文件、写入/修改文件 | 会按任务需要访问外部网络
安全层 -> 通常不需要额外 API Key + 小任务验证 + diff / 日志复核
输出层 -> 可复制结果 + 检查清单 + 下一步迭代
} Code Review
Use this skill when the user asks for a code review, second pass, pre-commit check, PR review, bug-risk review, or closeout review after non-trivial code edits.
This skill is for reviewing code. It may recommend fixes or apply small fixes when requested, but it should not become a broad refactor, redesign, or style-only cleanup pass.
Contract
- Treat review findings as hypotheses until verified against the real code path.
- Read the changed files, adjacent callers/callees, and relevant tests before judging a finding.
- Check dependency docs, source, or types when behavior depends on an external API or library.
- Prioritize correctness, regressions, security-sensitive behavior, data loss, race conditions, error handling, and test coverage.
- Avoid speculative edge cases, broad rewrites, taste-based style comments, or fixes that over-complicate the codebase.
- Prefer no finding over a weak or speculative finding.
- Prefer small, local fixes at the right ownership boundary.
- Preserve existing project conventions unless they are part of the problem.
- Do not push, commit, or open a PR unless the user explicitly asks.
- Lead final responses with findings, ordered by severity, using concrete file and line references when possible.
Pick Review Target
Choose the smallest target that matches the user's request.
Dirty local changes
Use when files are staged, unstaged, or untracked:
git status --short
git diff --stat
git diff
Include staged changes when relevant:
git diff --cached
Branch or PR changes
Use when reviewing a branch against its base:
git diff --stat origin/main...HEAD
git diff origin/main...HEAD
If a PR exists, use its actual base when available:
base=$(gh pr view --json baseRefName --jq .baseRefName)
git diff --stat "origin/$base...HEAD"
git diff "origin/$base...HEAD"
Fetch remote refs only when needed for an accurate branch/PR review. If network access is unavailable or fetching is inappropriate, use local refs and state that limitation.
Single commit
Use when reviewing one committed change:
git show --stat --oneline HEAD
git show HEAD
Workflow
- Identify the review target, user constraints, and whether fixes are requested.
- Inspect the diff first, then inspect surrounding code needed to understand each changed path.
- Map changed behavior to callers, state, errors, permissions, data flow, and tests.
- Verify each potential issue before reporting it.
- Check whether existing or missing tests would catch the changed behavior.
- Reject findings that are intentional, already covered, not reachable, or not worth the added complexity.
- If making fixes, keep them minimal and rerun the most focused validation.
- Stop when there are no accepted/actionable findings for the requested target.
Review Checklist
Focus on the areas that apply to the change:
- Correctness: logic errors, broken invariants, off-by-one issues, bad assumptions, incomplete state updates.
- Regressions: changed public behavior, compatibility breaks, migration risks, missing fallbacks.
- Error handling: swallowed errors, misleading messages, unsafe retries, missing cleanup.
- Security-sensitive behavior: auth/permission checks, input validation, injection paths, secret handling, unsafe file/network access.
- Concurrency and state: races, stale caches, ordering issues, duplicate side effects, reentrancy where relevant.
- Data integrity: serialization, parsing, rounding, units, schema changes, persistence, destructive operations.
- Tests: missing coverage for changed behavior, weak assertions, untested failure paths, brittle snapshots.
- Maintainability: confusing ownership boundaries, unnecessary coupling, duplicated complex logic.
Validation
Run the narrowest useful checks for the reviewed area. Prefer existing project commands from package.json, Makefile, CI config, README, or AGENTS.md.
Common examples:
npm test
npm run lint
npm run typecheck
pnpm test
pytest
cargo test
go test ./...
forge test
If validation is expensive, unavailable, or unrelated to the change, state what was skipped and why.
External Review Tools
If the environment provides a dedicated review command, use it as an advisory pass only after understanding the diff yourself.
Examples:
codex review --uncommitted
codex review --base origin/main
codex review --commit HEAD
Do not blindly apply tool findings. Verify each one in the code, accept only actionable findings, and rerun focused validation after any review-triggered fix.
Output
Final response should start with findings. Use this shape:
- Accepted findings ordered by severity. Include
path:line, impact, and the smallest useful fix direction. - If there are no actionable findings, say that directly.
- Review target inspected.
- Fixes made, if any.
- Validation run, or why validation was skipped.
- Remaining risks or assumptions.
Do not lead with a summary before findings. Mention rejected or intentionally ignored findings only when useful to explain judgment.
Source: Alacriity1/agent-skills — distributed by TomeVault.
先判断是否适合
作者设计意图
作者的方法与取舍
边界和复核