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 skills add XRenSiu/claude-code-forge --skill pr-reviewgit clone --depth 1 https://github.com/XRenSiu/claude-code-forgeWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/xrensiu/claude-code-forge/pr-review)<a href="https://agentmods.dev/skills/xrensiu/claude-code-forge/pr-review"><img src="https://agentmods.dev/badge/skills/xrensiu/claude-code-forge/pr-review/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/xrensiu/claude-code-forge/pr-review"><img src="https://agentmods.dev/badge/skills/xrensiu/claude-code-forge/pr-review.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What 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 5.1 | $0.00273 | $0.02707 |
| Opus 5.5 | $0.00109 | $0.01083 |
| Sonnet 5.5 | $0.00055 | $0.00541 |
| Haiku 4.5 | $0.00027 | $0.00271 |
Grade A, and why
pr-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 1mo ago.
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 — 114 lines — stays where its author put it; the contents beside it link to each section on GitHub.
pr-review — diff 进,能落地的发现出
产物是 findings.yaml(形状:assets/findings_template.yaml),可选地发成一个 GitHub review。本文件
写:发现的判据、三档映射、原语与出口、发之前的门。先读哪个 hunk、要不要多跳几层是你的份额。
缺口(Judgment + Capability + Control)
deletion 测试:撤掉本 skill,引擎读完 diff 回"整体结构清晰,有几个小建议"——没有一条带复现场景, 没有一条能直接改,reviewer 的时间被 nit 淹没,真正的 P0 被"看起来不错"盖住。缺的是判据(什么算 发现、什么不算)、档位(谁一票否决、谁只告警)、原语(把发现发到正确的行)与门(永不 approve)。
世界(Σ)
- 你看到的是 diff,不是整个仓库:diff 引用了你看不到的定义 →
LOCATE+READ去看,不因 "看不到"标发现;预算内看不到 →needs_codebase_check: true,交人。 - 三档(Spec Loop L7 的裁决 C9):A 机械档——测试 / lint / 类型 / secrets / 白名单 / 锁 / 新增依赖(注册表存在性、许可证、漏洞)/ 契约硬命中(改测试、删断言、mock 越界)——一票否决 = P0,有则一起修;B 结构档——圈复杂度 / 重复 / 公共 API 变更 / diff 体量 / 依赖方向——超阈值告警 = P1/P2,有界可进;C 判断档——架构意图、可读性、命名、human AC——只请求人工 = P3 或路由 G3。 一条发现只归一档,档位由内容决定不由语气决定。
- 发现的成本不对称:P0/P1 漏报 = 事故,误报 = reviewer 花 5 分钟;P2/P3 误报 = 噪声淹没信号。
所以 P0/P1 偏召回(中等置信也报),P2/P3 偏精确(不确定就不报,不报成
confidence: low)。 - 同模型审同模型有系统盲区:对抗式审查("假设这段代码造成了线上事故,找为什么")用另一家
模型(Codex / Gemini CLI 在则用)或至少不同尺寸;做不到就在输出顶部标
single-vendor caveat。 - GitHub review 的性质:行内评论只能挂在 diff 里出现的行(新文件坐标);不在 diff 里的位置降级
为 PR 级评论并引用
path:line;REQUEST_CHANGES会阻塞合并(有 A 档发现时才用); APPROVE 是人类动作。 - 关于用户的 Σ:"看看有没有问题"= 要发现,不要夸奖;"帮我 review 一下再合"= 要一个能否合的 结论 + 阻塞项清单,不是散文。
判据(φ)
- 一条发现 = 位置 + 机制 + 复现 + 证据 + 改法:
file:line_range、root_cause(一句机械描述)、 P0/P1 必有reproduction_scenario(具体输入 / 时序 / 状态),evidence(读了哪、grep 了什么、 git log 看到什么),suggested_change(P0/P1 必有,一行可执行)。缺任一项的 P0/P1 降级或删。 - 自我反驳后才发:把发现写成假设,"如果它不是 bug,代码会长什么样?实际代码是不是那样?"能 构造出合理的非 bug 解释 → 删。
- 上限 5 条(
--max-findings):超过按严重度再按证据强度取前 5;剩下的留给下一轮。发现少于 5 不凑;发现为 0 →findings: []+rationale(走了哪些路径、为什么没问题)。 - 一次一个 focus:
--focus=security只报安全类;其他焦点另开一次(all只在 Opus 级模型上用)。 - 禁语:
看起来不错/整体结构清晰/可以考虑/作为小建议——要么有证据的发现,要么没有。 - B 档阈值(可被
--rules覆盖):函数 > 50 行、嵌套 > 4、文件 > 500 行、重复块 > 20 行 ×2、 公共 API 签名变更无 CHANGELOG / 迁移说明、diff > 500 行。 - 新增依赖进 A 档:
package.json/requirements.txt/go.mod新增包 → 注册表存在(防编造包名)- 许可证 + 已知漏洞(
npm audit/pip-audit有则跑);查不到 → P0needs_codebase_check。
- 许可证 + 已知漏洞(
- 残差(人):架构意图对不对、这个抽象值不值——写成 C 档 P3,或直接建议 G3。
What ships with it
9 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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.
- 1mo ago Changed 810a195d7f3a
- 1mo ago First seen · 114 lines · 273 tokens per session scan A 789674d274e7
pr-review is a skill published in the GitHub repository XRenSiu/claude-code-forge (2 stars, last pushed 1mo ago), licensed MIT. It adds 273 tokens to every session and 2,707 once invoked, about $0.0011 per session on Opus 5.5. 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 2026-09-05.
Other skills, from other repositories
sync-bridge
Reviews and guides Chromium Sync Model API (DataTypeSyncBridge) implementations, including DataTypeStore persistence and forwards/backwards compatibility for sync specifics. Use when implementing or reviewing a DataTypeSyncBridge subclass, DataTypeStore usage, sync metadata persistence, conflict resolution, or…
sync-model-processor
Reviews changes to Chromium Sync model internals in components/sync/model/, including ClientTagBasedDataTypeProcessor, ClientTagBasedRemoteUpdateHandler, ProcessorEntity, ProcessorEntityMetadata, and ProcessorEntityTracker. Use when modifying or reviewing processor state transitions, entity metadata tracking, conflict…
adversarial-reviewer
Stress-test a code change for concrete correctness defects, unsafe assumptions, and failure modes. Use when the user explicitly asks for an adversarial review, a deep bug hunt, or to tear a change apart, or to check your own change before submitting it. Do not use for ordinary review, style feedback, or general…
gsd-ns-review
Invoke this Pi skill with /skill:gsd-ns-review.
gitnexus
A code-graph analysis add-on for examining an existing codebase, including symbols, call paths, execution flows, and effects across repositories. It can query GitNexus through its command-line or MCP interfaces.
issue
Use when starting a chain from a GitHub issue — turning an issue URL or number into a triaged, planned, dispatched, and reviewed pull request. Classifies the thread (bug → root-cause discipline, feature → plan chain, question → drafted reply), synthesizes a spec from the issue's own acceptance criteria, then runs the…