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 kopp0510/claude-dd --skill plan-reviewgit clone --depth 1 https://github.com/kopp0510/claude-ddWrote 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/kopp0510/claude-dd/plan-review)<a href="https://agentmods.dev/skills/kopp0510/claude-dd/plan-review"><img src="https://agentmods.dev/badge/skills/kopp0510/claude-dd/plan-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/kopp0510/claude-dd/plan-review"><img src="https://agentmods.dev/badge/skills/kopp0510/claude-dd/plan-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 | $0.00182 | $0.06859 |
| Opus 5.5 | $0.00073 | $0.02744 |
| Sonnet 5.5 | $0.00036 | $0.01372 |
| Haiku 4.5 | $0.00018 | $0.00686 |
Grade A, and why
plan-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 4d 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 — 236 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Plan Review — 裁決者 + 審查者的計畫審查
計畫做好後不直接交出去,先交給一個腦袋全新的裁決者。裁決者平行派審查者去證明計畫有重大缺失, 收回之後反過來查證每一條批評,只把站得住的交回來。主對話照著改,改到沒有站得住的重大缺失為止。
設計來源:docs/designs/2026-10-23-裁決審查-design.md(claude-dd repo)。
什麼時候跑
| 情境 | 跑不跑 |
|---|---|
全域 §4.1 的複雜任務(N = 1;N ≥ 2 等 task-planner 出草稿時再跑),計畫寫好、要交出去之前 |
跑 |
task-planner 步驟 4,草稿(進度表、小任務、測試清單)出好之後 |
跑 |
design-brainstorm 設計逐段整理好、離開 plan mode 之前 |
跑 |
| 使用者說「審計畫」「審查這個計畫」 | 跑 |
| 簡單、中等任務 | 不跑(但走 design-brainstorm 的設計不論大小都跑) |
| 審程式碼 diff | 不跑(那是步驟 4 的 code-review) |
步驟
1. 準備交給裁決者的東西
- 審查範圍:這次要審的是什麼(整份計畫,或某份文件的某一段)。只審一段時寫明「其他段落是背景,不在範圍內」, 這句同時寫進「事實」開頭(「審查範圍」那一欄審查者看不到)
- 計畫全文:照要交給使用者的版本,一字不改。計畫引用或依賴別處的段落(「見設計決策」、上層的設計決策表、 它依賴的段落),把那幾段一起附上,標「附錄:<出處>」;整份大文件放不下時,附路徑讓審查者自己讀
- 已查到的事實:讀過哪些檔(路徑)、實跑過的指令與輸出、使用者在對話裡講定的決定(逐條,標「使用者決定」)
- 只交事實,不交「我判斷 X 是原因」。要寫推測就標明是推測 —— 裁決者與審查者會跟著錯的框架走
- 第 2 輪以後:再附上前幾輪的裁決報告全文、「這輪改了哪裡」的清單、前幾輪外部查證的結果、前幾輪還沒問使用者的「需使用者決定」
- 輪次:這是第幾輪(從 1 開始)
- 專案根目錄的絕對路徑(審查者要自己 grep)
2. 派裁決者
用 Agent 工具:subagent_type: "general-purpose"、model: "opus",prompt 用下面的「裁決者範本」。
主對話只填 {…}(範圍、計畫、事實、輪次、根目錄);範本裡的 《…》 是留給裁決者填的,原樣保留。
計畫與事實放在 <<<…開始>>> / <<<…結束>>> 標記之間,不要用 Markdown 圍欄包(計畫裡有三個反引號會把圍欄提早關掉)。
裁決者在背景跑,要等它自己派的審查者,通常要好幾分鐘。等它的完成通知回來,不要自己先審、也不要先往下做。 沒有逾時機制:使用者問進度就照實說「還在跑」;使用者說不等了,就當成沒跑成(第 5 步)。
3. 照報告的「結論」走
結論只有三種值,照字面判,不要從別節推:
| 結論 | 主對話做什麼 |
|---|---|
通過 |
走第 4 步 |
不通過(N 條重大) |
見下方 |
無法判定(<原因>) |
當成沒跑成,走第 5 步 |
報告缺「結論」、或自相矛盾(寫 通過 但「採納(重大)」不是空的,或「需要外部查證」裡有標重大的)→ 當成沒跑成,走第 5 步。
不通過時:
- 「需使用者決定」裡有標了「卡住哪條重大」的題目 → 先用 AskUserQuestion 問這幾題;其他題留到下一輪的事實裡
- 照「採納(重大)」逐條改計畫;「需要外部查證」裡標重大的,主對話自己查(自己有工具就查,查不了就問使用者), 結果寫進下一輪的事實
- 「採納(小事)」順手改掉,不必為它再送一輪
- 改好的全文回到第 1 步,輪次加 1,再送一次
最多 3 輪。第 3 輪還是 不通過:重大與小事一樣照上面 1–3 改進計畫,然後停下來回報,不要自己送第 4 輪、也不要自己開工。
沒有下一輪了,所以 1 的「其他題」留到下面停下時一起問;2 的外部查證查到屬實就照改(未經審查),查到不屬實寫「不成立(查證結果)」,自己查不了、問了使用者也沒有結果的寫「沒改(外部查證還沒結果)」。
回報時每條重大寫明現況:「已改進計畫(未經審查)」或「沒改(原因)」——「沒改」只用在使用者決定不改、或外部查證還沒結果的那幾條(查證不屬實寫「不成立」);
只寫「剩 N 條重大」會被讀成還沒修。所有還沒問的「需使用者決定」一起問,再問接下來怎麼做:
跑一次確認輪(建議:第 3 輪的改法動到計畫內容時)、照改好的計畫繼續、或停下。
選照改好的計畫繼續 → 照第 4 步第 2、3 項做,最終計畫標明哪些改法未經審查(第 3 輪與確認輪改的重大與小事、因使用者答案改的地方)。
選停下(這裡或下面確認輪那幾條)→ 貼出改好的計畫與每條重大的現況,然後結束,不開工;呼叫端另有寫法時照呼叫端。
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.
- 4d ago Changed · +17 lines 8990685e226d
- 5d ago First seen · 219 lines · 182 tokens per session scan A 7d4fee9b9b67
plan-review is a skill published in the GitHub repository kopp0510/claude-dd (12 stars, last pushed 4d ago), licensed MIT. It adds 182 tokens to every session and 6,859 once invoked, about $0.0007 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 2026-10-25.
Other skills, from other repositories
autoreview
Structured code review through the shared OpenClaw agent-skills installation.
revdiff-plan
Review the last Codex assistant message (plan, analysis, or proposal) with inline annotations in a TUI overlay. Extracts the most recent response from Codex rollout files and opens it in revdiff for review and annotation. Activates on "revdiff-plan", "review plan with revdiff", "annotate plan", "review last response"…
re0-merge
Review and land an external contribution the way this suite does: gate it against the thesis, land it with the author's credit intact, complete a new skill rather than merging it raw, then approve, credit, and explain before closing. Use when reviewing a pull request, as any collaborator or maintainer, not only the…
asplos-writing-style
Use when drafting or revising ASPLOS prose — engineering the first two pages that rapid review actually reads, stating the cross-layer insight as one quotable sentence, structuring per-layer mechanism sections, writing bounded claims, and fitting everything into 11 self-contained pages that include figures and…
ring:mapping-feature-relationships
Mapping how features relate and phasing the work: categorizing PRD features, grouping them into domains, charting cross-feature journeys, dependencies, and integration points, and defining the binding Phases that plan.md mirrors one-to-one at Gate 7. Gate 2 of ring:planning-large-features; runs after…
code-reviewer
Code review specialist focused on patterns, bugs, security, and performance.