@mingfuyun/star-allinone 0.4.17 → 0.4.18

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@mingfuyun/star-allinone",
-  "version": "0.4.17",
+  "version": "0.4.18",
   "description": "星系运维前端能力包",
   "main": "dist/index.js",
   "bin": {

package/skills/shared/dev-flow/SKILL.md

@@ -19,7 +19,7 @@ disable-model-invocation: true
 
 - 提交、发布：不要打开本 skill 的 references。直接执行对应 skill，用户可见回复与单点相同。
 - 需求或接口调整：只在出方案时读 `plan-template.md`。`确认方案` 之后按清单连续执行原 skill。
-- **禁止**为了填 `phase` / `done` / `blocked` / `next_action` 单独停一轮，也禁止在两个 skill 之间插入「分析影响面 / 进入某阶段」的回复。
+- **禁止**为了填 `phase` / `done` / `blocked` / `next_action` 单独停一轮，也禁止在两个 skill 之间另插「进入某阶段」的旁白。`await_commit` 的用户可见正文是 `shared-review-and-submit` 的审查报告（影响面、审查、修改范围），这不是插入的分析轮，禁止摘要掉。
 - 契约分支能从用户原话或已有 diff 看出来时，不要读 `swagger-gate.md` 再分析一遍；只有不确定时才问一句。
 - `session-protocol.md`、`gates.md` 不必每轮 Read。口令以本文件和被委托 skill 为准。
 
@@ -111,7 +111,7 @@ next_action: …
 
 ### implement
 
-按方案里的有序清单连续执行原来的 skill。禁止跳步、禁止扩大清单外范围。每个原子 skill 完整 Read 后原样执行。两个 skill 之间不要停下来做分析或汇报阶段。用户看见的就是该 skill 自己的输出。
+按方案里的有序清单连续执行原来的 skill。禁止跳步、禁止扩大清单外范围。每个原子 skill 完整 Read 后原样执行。实现和校验的中间过程不要另写分析汇报。进入 `await_commit` 后，下一条用户可见回复就是审查报告。
 
 ### verify
 
@@ -123,7 +123,13 @@ next_action: …
 
 ### await_commit
 
-Read 并原样执行 `.agents/skills/shared-review-and-submit/SKILL.md`。停住等待 `确认提交` 或 `确认提交并推送`。口令以该 skill 与 [references/gates.md](references/gates.md) 为准。
+Read 并原样执行 `.agents/skills/shared-review-and-submit/SKILL.md`。停住前的那条用户可见回复必须带齐该 skill 的三节，不要另写清单，也不要收成摘要：
+
+1. **影响面**：`references/impact-analysis.md`
+2. **审查**：`references/verify.md` 与 `references/ai-review.md`
+3. **修改范围**：`references/commit.md`
+
+改动只有一行也要写全。然后停住，等待 `确认提交` 或 `确认提交并推送`。口令以该 skill 与 [references/gates.md](references/gates.md) 为准。不在这条回复里执行 git 写入。
 
 提交完成后：
 

package/skills/shared/i-have-adhd/i-have-adhd.md

@@ -96,6 +96,8 @@ For long lists in the final response, group related items and rank the most rele
 
 Never omit relevant items when completeness matters. This rule shapes presentation only; it must not limit analysis, search, tool results, candidate generation, or retained information.
 
+The pre-submit review report is one of those cases. Keep the sections 影响面, 审查, and 修改范围 on screen. Do not hold them back under the five-item cap.
+
 ### 10. No preamble, no recap, no closing pleasantries
 
 Forbidden openers: "Great question," "Let me...", "I'll...", "Sure!", "Looking at your...", "To answer your question..."
@@ -106,6 +108,8 @@ Forbidden closers: "Let me know if you need anything else," "Hope this helps," "
 
 Start with the answer. End when the answer is done.
 
+The review report is the answer for a commit wait. It is not a recap of the implementation, so do not delete it in the pre-send check.
+
 ## When to break the rules
 
 Override the defaults when:
@@ -116,6 +120,7 @@ Override the defaults when:
 4. Real ambiguity in the request. One short clarifying question beats guessing and rewriting.
 5. A rule fights the task. When a rule would delete the answer itself, the task wins; the shape stays. Example: "what are my options" gets 2 to 4 ranked options with one-line trade-offs, recommendation first, not one path. The options are the answer.
 6. A rule fights the harness. Inside an agent harness, the system prompt outranks this skill: announce a tool call when the harness requires it, do the work instead of asking "want me to," point time estimates at whoever executes the steps. Same principle as 5: the constraint wins, the shape stays.
+7. The message waits for `确认提交` or `确认提交并推送`. Output the full `shared-review-and-submit` report: 影响面, 审查, 修改范围. The first line can be the confirmation phrase. Do not collapse those sections.
 
 ## Pre-send check
 

package/skills/shared/review-and-submit/SKILL.md

@@ -73,3 +73,11 @@ If the user says `提交代码` before a review proposal exists, run the three s
 ## User-facing output
 
 Read and apply `.agents/skills/shared-i-have-adhd/i-have-adhd.md` to every response this skill sends to the user.
+
+The message that waits for `确认提交` or `确认提交并推送` is the review report. Reuse the three workflow steps above. Do not write a second checklist, and do not replace the report with `未发现阻断问题` plus a commit subject. A one-line diff does not shorten it. ADHD mode may put the confirmation phrase on the first line. It must not delete these sections:
+
+1. **影响面** — from [references/impact-analysis.md](references/impact-analysis.md): what changed, callers and pages, related APIs / types / permissions / routes, shared-component blast radius, and how to regress.
+2. **审查** — from [references/verify.md](references/verify.md) and [references/ai-review.md](references/ai-review.md): P0–P2, tracking / permission / async / empty states, command results, uncovered risks, plus hooks, API contract, permission, and forms.
+3. **修改范围** — from [references/commit.md](references/commit.md): findings, validation, branch and remote, exact file paths, and the commit message.
+
+Then stop. Do not run Git writes in that message.