ask-later 0.1.0 → 0.1.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/.claude-plugin/plugin.json +1 -1
- package/README.en.md +8 -5
- package/README.md +9 -5
- package/package.json +7 -3
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "ask-later",
|
|
3
|
-
"version": "0.1.
|
|
3
|
+
"version": "0.1.2",
|
|
4
4
|
"description": "Ask now, get the answer later. Shelves the agent's questions in PENDING.md, injects the human's answers from INBOX.md on the next tool call, and blocks the stop (twice by default) while unread lines remain. For Claude Code; unofficial.",
|
|
5
5
|
"author": { "name": "metamol0627" },
|
|
6
6
|
"license": "MIT",
|
package/README.en.md
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
|
-
# ask-later —
|
|
1
|
+
# ask-later — run Claude Code unattended without stopping at questions (an async question queue)
|
|
2
2
|
|
|
3
|
-
Two files and five hooks that keep Claude Code from stopping at "may I?".
|
|
3
|
+
**Run Claude Code unattended: questions are queued, work continues.** Two files and five hooks that keep Claude Code from stopping at "may I?".
|
|
4
|
+
Tool permissions are a settings matter (allow lists, auto-approve); a question that needs a human's judgment is not, and that is what ask-later handles.
|
|
4
5
|
The agent files what it needs from a human in `PENDING.md`, shelves only the work that depends on the answer, and keeps going.
|
|
5
6
|
The human writes one line in `INBOX.md` with any text editor. A hook injects that line into the agent's context on its next tool call,
|
|
6
7
|
and blocks the stop while unread lines remain (twice by default). A sentinel file named `!pending_3.txt` at the project root means "3 open items".
|
|
@@ -9,6 +10,8 @@ Node.js 20+, zero dependencies, MIT, unofficial. **Claude Code only** — it spe
|
|
|
9
10
|
Events used: `SessionStart` / `UserPromptSubmit` / `PostToolUse` / `Stop` / `SessionEnd`.
|
|
10
11
|
Verified only on Windows 11, Node 24, Claude Code 2.1.268 (macOS/Linux not yet tried). The Japanese `README.md` is the primary document; this is a summary.
|
|
11
12
|
|
|
13
|
+
Sold separately: "無人運転の設計書" (a design guide for unattended runs, in Japanese, ¥1,480) — [sales page](https://lab-cws.pages.dev/ask-later/). The tool itself is free.
|
|
14
|
+
|
|
12
15
|
## Read this first (what gets written)
|
|
13
16
|
|
|
14
17
|
- `npx ask-later init` writes three things: **the `hooks` key of `.claude/settings.json`** (other keys and other people's hooks are left alone),
|
|
@@ -106,7 +109,7 @@ In an interactive bash, quote the sentinel name (`'!pending_1.txt'`) — history
|
|
|
106
109
|
|
|
107
110
|
## As a plugin
|
|
108
111
|
|
|
109
|
-
The package ships `.claude-plugin/plugin.json` and `hooks/hooks.json`. `npm pack ask-later && tar xf ask-later
|
|
112
|
+
The package ships `.claude-plugin/plugin.json` and `hooks/hooks.json`. `npm pack ask-later && tar xf ask-later-*.tgz && claude --plugin-dir ./package` enables the hooks without `init` (create the templates with `npx ask-later init --no-hooks`). Marketplace installation will be documented once a public repository exists.
|
|
110
113
|
|
|
111
114
|
## Known limits
|
|
112
115
|
|
|
@@ -114,11 +117,11 @@ Verified on Windows only. Whether the agent obeys the injection depends on the m
|
|
|
114
117
|
|
|
115
118
|
## Paid companion
|
|
116
119
|
|
|
117
|
-
The tool is free (MIT). A separate **"Unattended operation design book" (JPY 1,480, Japanese, Markdown + HTML ZIP, 7 chapters, 10 templates, ~300 KB)** covers the discipline rather than the tool: why agents stop, the shelving queue, four brakes for unattended runs, independent-session audits, machine-written records, a dated log of 14 real failures, and a template set. For people running `-p` or auto-resume unattended; not needed for interactive-only use. Chapter 1 and the minimal templates are free in this package under `docs/manual-free/` (open `index.html`).
|
|
120
|
+
The tool is free (MIT). A separate **"Unattended operation design book" (JPY 1,480, Japanese, Markdown + HTML ZIP, 7 chapters, 10 templates, ~300 KB)** covers the discipline rather than the tool: why agents stop, the shelving queue, four brakes for unattended runs, independent-session audits, machine-written records, a dated log of 14 real failures, and a template set. For people running `-p` or auto-resume unattended; not needed for interactive-only use. Chapter 1 and the minimal templates are free in this package under `docs/manual-free/` (open `index.html`). How to get it: on the [sales page](https://lab-cws.pages.dev/ask-later/), click "¥1,480 で購入する" (Buy for ¥1,480), pay through Stripe, and the confirmation screen takes you to a download page for the ZIP.
|
|
118
121
|
|
|
119
122
|
## Development
|
|
120
123
|
|
|
121
|
-
`npm test` (node --test), `npm run fieldlog` (runs the hooks on a fixed copy of the real files this tool was ported from and cross-checks counts against the original PowerShell implementation; see `docs/field-log.md`, Japanese), `npm run build:manual`.
|
|
124
|
+
`npm test` (node --test), `npm run fieldlog` (runs the hooks on a fixed copy of the real files this tool was ported from and cross-checks counts against the original PowerShell implementation; see `docs/field-log.md`, Japanese), `npm run build:manual`. `npm test` needs `tests/` (and part of `scripts/`), which are not in the tarball (it will run in the public repository once there is one). `fieldlog` and `build:manual` need the author's private data and the paid manuscript, so they will not run there either.
|
|
122
125
|
|
|
123
126
|
## License
|
|
124
127
|
|
package/README.md
CHANGED
|
@@ -1,6 +1,7 @@
|
|
|
1
|
-
# ask-later —
|
|
1
|
+
# ask-later — Claude Code が質問で止まらない(聞いて、待たずに進む・非同期の質問キュー)
|
|
2
2
|
|
|
3
|
-
**Claude Code
|
|
3
|
+
**Claude Code を無人で回すと「確認していいですか」で止まります。その質問を後回しにして、作業を続けさせる——ファイル 2 つとフック 5 つ**です。
|
|
4
|
+
ツールの許可(許可リストや自動承認)は設定で決まりますが、人の判断が要る「質問」は設定では答えられません。ask-later が扱うのはこちらです。
|
|
4
5
|
エージェントは人に聞きたいことを `PENDING.md` に起票して、それに依存する作業だけを止め、残りを続けます。
|
|
5
6
|
人はメモ帳で `INBOX.md` に 1 行書くだけ。フックが次のツール呼び出しでその行をエージェントの文脈に注入し、
|
|
6
7
|
未読が残ったまま終わろうとすると差し止めます(既定 2 回)。プロジェクトのルートに置かれる `!pending_3.txt` という名前のファイルが「未処理 3 件」を示します。
|
|
@@ -9,6 +10,8 @@ Node.js 20 以上・依存パッケージなし・MIT・非公式。**Claude Cod
|
|
|
9
10
|
使うイベントは `SessionStart` / `UserPromptSubmit` / `PostToolUse` / `Stop` / `SessionEnd`。
|
|
10
11
|
動作確認は Windows 11・Node 24・Claude Code 2.1.268 のみ(macOS/Linux は未確認)。
|
|
11
12
|
|
|
13
|
+
別売りで「無人運転の設計書」(¥1,480)があります → [販売ページ](https://lab-cws.pages.dev/ask-later/)(詳しくは末尾「設計書」)。この道具そのものは無料です。
|
|
14
|
+
|
|
12
15
|
> **先に読んでください(何を書き換えるか)**
|
|
13
16
|
>
|
|
14
17
|
> - `npx ask-later init` が書き換えるのは **`.claude/settings.json` の `hooks`**(他のキーは触りません。既にあるフックも残します)、
|
|
@@ -250,7 +253,7 @@ npx ask-later --version / --help
|
|
|
250
253
|
同じパッケージに `.claude-plugin/plugin.json` と `hooks/hooks.json` が入っています。展開したディレクトリを指定して起動すると、`init` を使わずにフックが有効になります(この形では `PENDING.md`・`INBOX.md` は自分で置くか、`npx ask-later init --no-hooks` で雛形だけ作ります)。
|
|
251
254
|
|
|
252
255
|
```
|
|
253
|
-
npm pack ask-later && tar xf ask-later
|
|
256
|
+
npm pack ask-later && tar xf ask-later-*.tgz
|
|
254
257
|
claude --plugin-dir ./package
|
|
255
258
|
```
|
|
256
259
|
|
|
@@ -287,12 +290,13 @@ claude --plugin-dir ./package
|
|
|
287
290
|
対象は `-p` や自動再開で無人運転をしている人。対話でしか使わない人には要りません。
|
|
288
291
|
|
|
289
292
|
**第 1 章と最小テンプレは無料で、このパッケージの `docs/manual-free/` に入っています**(`npm pack ask-later` で取れます。`index.html` を開けば読めます)。
|
|
290
|
-
**入手方法**:
|
|
293
|
+
**入手方法**: [販売ページ](https://lab-cws.pages.dev/ask-later/)の「¥1,480 で購入する」から Stripe で決済すると、決済完了の画面から ZIP のダウンロードページに移ります。届かない場合の連絡先は販売ページの「特定商取引法に基づく表記」にあります。
|
|
291
294
|
|
|
292
295
|
## 開発者向け
|
|
293
296
|
|
|
294
297
|
ソースコードは npm のパッケージに入っています(`npm pack ask-later`)。公開リポジトリは準備中です。
|
|
295
|
-
下のコマンドのうち `npm test
|
|
298
|
+
下のコマンドのうち `npm test` は tarball に入っていない `tests/`(と `scripts/` の一部)が要ります(公開リポジトリができたらそちらで動きます)。
|
|
299
|
+
`fieldlog` と `build:manual` は作者の手元にある実データと有償版の原稿が要るので、公開リポジトリでも動きません。
|
|
296
300
|
|
|
297
301
|
```
|
|
298
302
|
npm test # 自動テスト(node --test。PENDING の解析・INBOX の未読・注入の重複抑止・Stop の差し止め・init/uninit の往復・CLI)
|
package/package.json
CHANGED
|
@@ -1,7 +1,9 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "ask-later",
|
|
3
|
-
"version": "0.1.
|
|
4
|
-
"
|
|
3
|
+
"version": "0.1.2",
|
|
4
|
+
"homepage": "https://lab-cws.pages.dev/ask-later/",
|
|
5
|
+
"_homepageNote": "npm のパッケージページの右側に出る唯一のリンク欄。公開リポジトリは未作成(404 を出さないので repository/bugs は入れない)。0.1.0 の README には販売ページの URL が 1 箇所も無く、npm から購入ページへ渡る橋が存在しなかった(判断監査 J-SHIP-BOTTLENECK 指摘1・2026-09-30)。",
|
|
6
|
+
"description": "Run Claude Code unattended without stopping at questions: the agent shelves questions in PENDING.md and keeps working, your one-line answers in INBOX.md are injected on the next tool call, and the stop is blocked (twice by default) while lines are unread. — Claude Code を無人で回すと「確認していいですか」で止まる——その質問を後回しにして作業を続ける。Claude Code 用の非同期の質問キュー(非公式)。",
|
|
5
7
|
"keywords": [
|
|
6
8
|
"claude-code",
|
|
7
9
|
"hooks",
|
|
@@ -47,7 +49,9 @@
|
|
|
47
49
|
"scripts": {
|
|
48
50
|
"test": "node --test tests/*.test.mjs",
|
|
49
51
|
"fieldlog": "node scripts/field_log.mjs",
|
|
50
|
-
"build:manual": "node scripts/build_manual.mjs"
|
|
52
|
+
"build:manual": "node scripts/build_manual.mjs",
|
|
53
|
+
"check:claims": "node scripts/check_claims.mjs",
|
|
54
|
+
"check:sale": "node ../../scripts/check_sale_status.mjs ask-later"
|
|
51
55
|
},
|
|
52
56
|
"dependencies": {}
|
|
53
57
|
}
|