Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx agentmods add rules/oldjii/code-review-mcp/code-reviewgit clone --depth 1 https://github.com/OldJii/code-review-mcpWhat it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5 | $0.00007 | $0.01937 |
| Opus 5 | $0.00003 | $0.00968 |
| Sonnet 5 | $0.00001 | $0.00387 |
| Haiku 4.5 | $0.00001 | $0.00194 |
Grade A, and why
code-review scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured yesterday.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 269 lines — stays where its author put it; the contents beside it link to each section on GitHub.
代码审查规范
根据用户输入语言自动选择规范版本。中文输入使用本文件,英文输入使用 code-review-en.mdc。
核心原则
理解优先
- 先理解 PR/MR 整体目的,再分析 diff
- 深度分析(架构/逻辑/性能/安全)优先于风格检查
- 只提"确认的问题",不提"可能的问题"
- 结合上下文分析,不孤立判断
审查流程
步骤 1:确定 PR 类型
主仓库 PR -> 从描述中提取子模块 PR
子模块 PR -> 直接审查
步骤 2:获取完整变更
changes = get_pr_changes(provider, repo, pr_id)
for change in changes["changes"]:
full_diff = change["diff"]
analyze(full_diff)
必须:遍历所有文件的完整 diff
禁止:diff[:100] 或 changes[:10]
步骤 3:理解 PR 上下文(优先执行)
在查看任何 diff 之前,回答:
- 目的(从标题+描述):新功能/Bug修复/重构?
- 范围(从变更列表):新增/删除/修改了哪些文件?
- 架构:类关系、数据流、设计模式?
- 预期实现:应该实现什么?
步骤 4:深度代码分析
4.1 理解 diff 意图
- 这段 diff 在整个 PR 中的作用?
- 是否符合 PR 目的?
4.2 架构审查
- 职责是否单一?
- 是否有重复代码?
- 抽象层次是否清晰?
- 设计在整体架构中是否合理?
4.3 逻辑审查(结合上下文)
- 边界条件:调用方是否已确保非空?
- 异常场景:业务流程中是否真的会发生?
- 状态转换:在完整状态机中是否完整?
避免误报:调用方已校验 -> 不提
4.4 性能审查(真实场景)
- 数据量多大?是否真的会成为瓶颈?
- 循环次数?是否真的影响性能?
- 耗时多久?是否真的会卡顿?
避免误报:数据量 <= 5 -> 不提 / 耗时 < 1ms -> 不提
4.5 安全与崩溃风险
- 确认真实的 NPE/越界/资源泄漏风险
- 考虑完整调用链和业务流程
避免误报:上文已校验 -> 不提
步骤 5:去重
# 1. 去重:同一文件相同问题合并
seen = {(file, issue_type): comment}
merge_line_numbers(seen[key], comment)
# 2. 按优先级分组
p0, p1, p2 = group_by_priority(seen.values())
# 3. 动态调整
if len(p0) > 10:
p1 = p1[:len(p1)//2]
p2 = p2[:len(p2)//3]
final_comments = p0 + p1 + p2
去重规则:
- 同一文件相同问题 -> 合并并列出所有行号
- 不同文件相同问题 -> 每个文件 1 条
- 方法过长 -> 每个文件最多 2 条
步骤 6:展示评论供确认
提交前必须展示所有评论供用户确认
格式:
## 审查结果预览
### 行内评论 (N 条)
1. [P0/P1/P2] 文件名:行号
评论预览...
2. [P0/P1/P2] 文件名:行号
评论预览...
### 总结评论
[总结预览]
---
等待确认后提交。
步骤 7:确认后提交
重要:使用 line + 1 避免遮挡代码行
for comment in inline_comments:
comment['line'] = comment['line'] + 1
batch_add_comments(
provider=provider,
repo=repo,
pr_id=pr_id,
inline_comments=inline_comments,
pr_comment=summary_comment
)
评论格式
行内评论
数量策略:
- P0:不限
- P1:P0 > 10 时减少 50%
- P2:P0 > 10 时减少 70%
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- yesterday First seen · 269 lines · 7 tokens per session scan A bbb8ddaef385
code-review is a cursor rule published in the GitHub repository OldJii/code-review-mcp (4 stars, last pushed 6d ago), licensed MIT. It adds 7 tokens to every session and 1,937 once invoked, about $0.0000 per session on Opus 5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-31.
Other cursor rules, from other repositories
archcore-files
Enforce MCP-only operations when working with .archcore/ files.
dev_workflow
Guide for using Taskmaster to manage task-driven development workflows.
execution
Repository execution and verification commands.
cosmosmith
Cosmosmith project rules adapter.
no-crux-frontmatter
This is a regular rule file that does not have crux: true in its frontmatter. It should be ignored by the CRUX compression hook.
fix-issue
Implement a fix following the human-thinking loop — understand the root cause, plan the minimal change, implement, verify the problem is actually gone.