adhd-review

v0.6.0

回覆先講行動，阻礙排在參考資訊之前

一套規則，分成三層。第 1 層調整每一則回覆。回覆先寫行動，多步驟工作加上編號與時間估計。回覆刪除開場白、重述與結尾，錯誤只陳述事實。第 2 層管理大型多步驟工作的最後一則回覆。這則回覆使用 Review-Ready 分類：✅ Done / ⚠️ Broken / 🙋 What I need from you / 🤖 What I'll do。阻礙排在參考資訊之前，每個請求都說明為何由你決定。視覺層把非線性的流程概念畫成 ASCII 圖，取代文字說明。圖放在程式碼區塊中。這類概念包括分支、迴圈、狀態變化與階層。線性步驟仍然用編號清單。規則只套用在面向使用者的主對話。預設開啟的 SessionStart hook 會把規則注入主工作階段。因此，子代理的回傳仍然保留完整細節。外掛預設開啟，安裝後主對話立即套用規則。如果要關閉某個工作階段，請設定 CLAUDE_ADHD_REVIEW=0。用 /adhd-review-mode 可以切換單一工作階段。

安裝

$ claude plugin install adhd-review@21-breakincode

Skill（自動啟用）

adhd-review-mode Use to toggle the adhd-review reply style for the current session.

設定

變數	預設值	說明
CLAUDE_ADHD_REVIEW	unset → on	Per-session kill switch. On by default — installing the plugin auto-applies the style to every new main session. Set to `0` to disable it for that session; unset (or `1`) re-enables. The style targets the main thread only; subagents never inherit it.

更新紀錄

0.6.0 2026-09-24

refactor ship the reply rules through the SessionStart hook only. The plugin no longer ships an output style, so you never set outputStyle for it. The rules moved from output-styles/adhd-review.md to scripts/session-rules.md. If your settings set outputStyle to adhd-review:ADHD Review, delete that line.
refactor remove the outputStyle check and the frontmatter strip from scripts/session-start.sh. With one delivery path, the rules cannot load twice.

0.5.6 2026-09-24

fix skip the SessionStart injection when outputStyle already selects ADHD Review. Claude Code loads that body natively, so the hook added a second copy (about 1,374 tokens) at every start, resume, clear, and compact.
fix set keep-coding-instructions: true on the output style. Without it, Claude Code removes its built-in software engineering instructions while the style is active.

0.5.5 2026-09-19

fix rewrite skills/adhd-review-mode/SKILL.md to clear the simple-english linter. No content or meaning changed, only sentence shape.

0.5.4 2026-09-19

fix rewrite the remaining prose (Scope guard, Layer 1, Visual Layer, Layer 2) to clear the simple-english linter. No content or meaning changed, only sentence shape.
feat add a hard rule to the Visual Layer. Diagrams in a reply are always ASCII. Never use mermaid, since a plain chat reply cannot render it as a picture.

0.5.3 2026-09-19

fix remove the Language Layer (the embedded ASD-STE100 Simple English rules). This plugin now covers reply shaping only. Install the simple-english plugin for language rules. That plugin already covers this job.

0.5.2 2026-09-13

fix strengthen diff instruction with anti-pattern catalog, positive example, and defense-in-depth reinforcement

0.5.1 2026-09-13

fix code-diff format picker rule now demands verbatim code lines, not intent summaries

0.5.0 2026-08-28

feat add Language Layer that enforces ASD-STE100 Simplified Technical English as mandatory on every reply (all 53 rules + 8 general recommendations)

0.4.0 2026-08-15

feat expand Visual Layer with technique vocabulary (pseudocode, call trees, component/file trees, mermaid, diffs, HTML artifacts)
feat add format picker for content-type to view mapping (sequence diagrams, diff view, flow diagrams, done/action view)

0.3.0 2026-07-29

feat add Visual Layer with fenced ASCII diagrams for flow-shaped concepts

0.2.0 2026-07-29

refactor default-on with per-session opt-out

0.1.0 2026-07-27

feat initial release of ADHD-friendly per-reply shaping + Review-Ready handoff summaries