Getting it into your agent
This one installs as part of its plugin. Adding the marketplace and installing the plugin brings it with everything else the plugin ships.
/plugin marketplace add alvis-HaoH/gkd/plugin install gkdWrote 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/commands/alvis-haoh/gkd/review)<a href="https://agentmods.dev/commands/alvis-haoh/gkd/review"><img src="https://agentmods.dev/badge/commands/alvis-haoh/gkd/review.svg" alt="Measured on agentmods" 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 | $0.00044 | $0.01181 |
| Opus 5 | $0.00022 | $0.00590 |
| Sonnet 5 | $0.00009 | $0.00236 |
| Haiku 4.5 | $0.00004 | $0.00118 |
Grade A, and why
review scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 8d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 55 lines — stays where its author put it; the contents beside it link to each section on GitHub.
让委派模型审查本地 git 改动。review-only:绝不修改文件、不给 patch、不暗示要改。 你只负责调度:定范围 → 选模式 → 委派 → 把子进程的审查输出逐字返回用户。审查指令本身在 prompt 模板里,由 runtime 注入子进程——你不用拼审查步骤。
这个命令是把审查工作委派出去:你做调度(定范围、选模式、回传),读 diff 和审代码是子进程的活——别让整个 diff 占满主上下文。
原始参数:$ARGUMENTS
1. 选 prompt 模板
- 默认
${CLAUDE_PLUGIN_ROOT}/prompts/review-defects.md(常规缺陷审查:bug/逻辑/边界/安全) - 用户传
--adversarial、或自然语言要求"挑战设计/质疑实现方式/对抗审" →${CLAUDE_PLUGIN_ROOT}/prompts/review-adversarial.md(对抗式设计审查) - 拿不准用户想要哪种就用
AskUserQuestion问用户**(选项:常规缺陷审查 / 对抗式设计审查)
2. 定范围(scope)
--scope working-tree:审未提交改动(已跟踪 + untracked)--scope branch:审<base>...HEAD整支(默认 base=origin/main,无远程则main,--base <ref>可指定)- 不传时
auto:git status --short --untracked-files=all非空 → working-tree,否则 → branch
3. 选执行模式
- 参数含
--wait→ 前台(必须timeout: 600000);含--background→ 后台;极小改动(1-2 文件、<50 行)推荐前台,其余(含估不清)推荐后台——审查易超 2 分钟,前台被杀代价大。拿捏不准时则估范围(git diff --shortstat等)后用AskUserQuestion问一次(Wait for results/Run in background,推荐项放首位)
4. 委派给 runtime
node "${CLAUDE_PLUGIN_ROOT}/scripts/gkd-runtime.mjs" \
<用户的 --<modelKey>(若有,否则不传走默认)> \
<用户给了 --effort <档> 就带上(none/low/medium/high/xhigh/max);审查是想深想的典型场景,用户说"仔细审/对抗审用 max"之类就补 --effort max> \
--allowed-tools "Read Grep Glob Bash(git:*)" \
--prompt-file "<上面选定的模板路径>" \
--render review \
"$(cat <<'__GKD_TASK_EOF__'
审查范围 <scope,如 working-tree 或 branch(base=origin/main)>。
<若用户给了额外关注点,粘在这里:重点关注 ...>
用 git 看改动(git status、git diff / git diff --cached、或 git diff <base>...HEAD,untracked 文件逐个读),然后按你的审查准则给意见。
__GKD_TASK_EOF__
)"
审查的立场/关注面/发现标准已在 --prompt-file 注入的模板里,任务文本只交代范围和额外关注点。--render review 让 runtime 把子进程返回的结构化 JSON 渲染成干净报告(模板已要求子进程只吐 JSON,过程独白被挡在输出之外)。前台传 timeout: 600000;后台用 run_in_background: true,报 task_id 给用户,收到 <task-notification> 后 Read .output。
输出处理
stdout 契约:runtime 的 stdout 就是给用户的最终产物,你是纯管道——逐字返回,不概括、不重写、不清洗、不根据它自行重建一份审查报告。 无论 stdout 是渲染好的干净报告(## 审查报告 + 分条发现 + 总判定),还是 [gkd] ⚠ 开头的降级提示(模型没吐合法 JSON 时的原文兜底),都照此逐字回传模型的最终判断,中间过程无需逐字返回。
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.
- 8d ago First seen · 55 lines · 44 tokens per session scan A 209ba49d481a
review is a command published in the GitHub repository alvis-HaoH/gkd (10 stars, last pushed 1mo ago), licensed MIT. It adds 44 tokens to every session and 1,181 once invoked, about $0.0002 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 commands, from other repositories
review
Structured code review with severity categorization (distinct from Claude Code's PR review).
discover
Run a full product discovery cycle — from outcome definition through opportunity mapping, prioritisation, and experiment design. Use when the team isn't sure what to build next, or before writing a PRD for a complex feature space.
simplify
The over-engineering review: five tags (delete, stdlib, native, yagni, shrink), a mandatory replacement per finding, and a real null result when there is nothing to cut.
git
The pre-finish status: branch, hygiene findings, message checks, workflow lint, template state.
init
Install the formatters this repository needs, with every command visible before it runs.
extend
Capture a mid-PR sub-idea and implement it onto the current open PR's branch — no new branch, no new PR. Preserves Verify → Review → Deliver.