Merge pull request #10 from jkrandom-sudo/feat/loop-dialog-ux

feat(tui): silent non-view commands; interactive task list dialog (0.5.0)

Author: jkrandom-sudo

AGENTS.md

@@ -0,0 +1,20 @@
+# AGENTS.md
+
+## Project
+
+opencode-plugin-loop — `/loop` command plugin for OpenCode (fixed / adaptive / maintenance scheduling).
+
+- Source: `src/` (TypeScript, SolidJS TUI views)
+- Tests: `tests/` (node:test, runs against the `dist/` build output)
+- Build + test: `npm test` (builds to `dist/` via Babel/tsc, then runs the full suite)
+
+## Development workflow (required)
+
+1. **Branch first**: cut a feature branch (e.g. `feat/xxx`) from the default branch; never commit directly to the default branch.
+2. **Verify locally**: `npm test` must pass; for TUI / command-behavior changes, also verify end-to-end in OpenCode.
+3. **PR and merge**: open a PR with `gh pr create` and merge into the default branch after verification.
+4. **Publish to npm**:
+   - Bump `package.json` semver after merging (feat → minor, fix → patch).
+   - The npm access token lives in `/Users/wangshuai/Downloads/npm_access_token.txt`; use the most recently issued token in that file.
+   - Use the token only via environment variables or throwaway publish-time config (e.g. `//registry.npmjs.org/:_authToken`); **never write it into the repo or commit it**; local `.npmrc` must stay out of git.
+   - Publish with `npm publish`, then confirm via `npm view opencode-plugin-loop version`.

CLAUDE.md

@@ -0,0 +1,20 @@
+# CLAUDE.md
+
+## 项目
+
+opencode-plugin-loop — OpenCode 的 `/loop` 命令插件，支持 fixed / adaptive / maintenance 三种调度模式。
+
+- 源码：`src/`（TypeScript，SolidJS TUI 视图）
+- 测试：`tests/`（node:test，运行于 `dist/` 构建产物之上）
+- 构建 + 测试：`npm test`（先 Babel/tsc 构建到 `dist/`，再跑全部测试）
+
+## 开发流程（必须遵守）
+
+1. **新分支开发**：所有改动从默认分支切 feature 分支（如 `feat/xxx`），不直接在默认分支上提交。
+2. **本地验证**：`npm test` 全部通过；涉及 TUI/命令行为的改动需在 OpenCode 中手动端到端验证。
+3. **提交 PR 并合并**：验证通过后 `gh pr create`，合并到默认分支。
+4. **发布 npm**：
+   - 合并后按语义化版本 bump `package.json` 版本号（feat → minor，fix → patch）。
+   - npm access token 存放在 `/Users/wangshuai/Downloads/npm_access_token.txt`，使用该文件中最新下发的 token 发布。
+   - token 只通过环境变量或发布时临时配置使用（如 `//registry.npmjs.org/:_authToken`），**绝不写入仓库、绝不提交**；本地 `.npmrc` 不得进入 git。
+   - 发布：`npm publish`，发布后 `npm view opencode-plugin-loop version` 确认。

README.md

@@ -29,10 +29,10 @@ A drop-in `/loop` command for [opencode](https://opencode.ai), modeled after Cla
 - **Auto-expire** — tasks idle for more than 7 days are removed on load (active tasks never expire)
 - **Max 50 concurrent tasks**
 - **LLM-callable tools** — `loop_schedule`, `loop_status` (session-bound by default)
-- **Interactive Loop results** — `/loop` results open in a dedicated native dialog instead of writing over the prompt
+- **Interactive task list** — `/loop list` and `/loop status` open a dedicated native dialog with selectable task rows instead of writing over the prompt; start/cancel/pause/resume stay silent, failures surface as error toasts
 - **Clipboard actions** — copy the complete result or copy any displayed task ID with one action
 - **Keyboard and mouse navigation** — move with `Up`/`Down` or `Tab`/`Shift+Tab`, hover with the pointer, and activate with `Enter`, `Space`, or a click
-- **Responsive layout** — short or narrow terminals keep the dialog inside the viewport with independently scrollable result and action areas
+- **Responsive layout** — short or narrow terminals keep the dialog inside the viewport with a scrollable task list
 - **Easy dismissal** — choose **Close**, press `q`, or use the native dialog's `Esc` key
 
 ## Requirements
@@ -187,15 +187,15 @@ If you try `cancel <id>` for a task owned by another session, you'll get a refus
 
 Two behavioral differences worth knowing: tasks only fire for the **currently active session** (switch sessions and the others wait; switch back and they catch up once), and fixed tasks fire on a 5-second ticker rather than exact wall-clock cron times (up to one ticker period late).
 
-### Interactive result dialog
+### Interactive task list dialog
 
-Every `/loop` command result opens in a separate native OpenCode dialog. It keeps task output away from the prompt and provides:
+`/loop list` and `/loop status` open a native OpenCode dialog rendering your tasks as a selectable, color-coded list (▶ active, ⏸ paused). Starting, cancelling, pausing, or resuming a task stays silent; failures surface as error toasts. The dialog provides:
 
-- **Copy ID: `<taskId>`** for every distinct task shown in the result
+- A highlighted **task row** per task — press `Enter` to copy its task ID
 - **Copy all** for the exact complete result text
 - **Close** to dismiss the dialog
 
-Use `Up`/`Down` or `Tab`/`Shift+Tab` to change the selected action, then press `Enter` or `Space` to activate it. Moving the mouse over a row selects it, and clicking activates that exact row. A successful **Copy ID** or **Copy all** action shows a confirmation and closes the dialog immediately; if clipboard access fails, the dialog stays open and shows an error. Press `Page Up` or `Page Down` to scroll long result text, or press `q` or `Esc` to close. In short or narrow terminals, the dialog scales to the available viewport and keeps the result and action lists independently scrollable. A newer Loop result replaces the previous Loop dialog rather than stacking another one.
+Use `Up`/`Down` or `Tab`/`Shift+Tab` to change the selected row, then press `Enter` or `Space` to activate it. Moving the mouse over a row selects it, and clicking activates that exact row. A successful copy shows a confirmation and closes the dialog immediately; if clipboard access fails, the dialog stays open and shows an error. Press `Page Up` or `Page Down` to scroll long lists, or press `q` or `Esc` to close. In short or narrow terminals, the dialog scales to the available viewport and keeps the list scrollable. A newer Loop result replaces the previous Loop dialog rather than stacking another one.
 
 ### Programmatic (LLM tools)
 

package-lock.json

@@ -1,6 +1,6 @@
 {
   "name": "opencode-plugin-loop",
-  "version": "0.4.0",
+  "version": "0.5.0",
   "description": "/loop command for opencode — run prompts on a schedule (fixed, adaptive, or maintenance), modeled after Claude Code's /loop",
   "type": "module",
   "main": "./dist/index.js",

package.json

@@ -71,13 +71,17 @@ export async function showLoopResult(
   result: CommandParseResult,
   logger: LoopLogger
 ): Promise<void> {
+  const variant = toastVariant(result.message)
+  // Non-view results (start/cancel/pause/resume/stop-all) stay silent by design;
+  // only task lists (info) and failures (error) surface a toast.
+  if (variant === "success") return
   try {
     await client.tui.showToast({
       throwOnError: true,
       body: {
         title: LOOP_FEEDBACK_TITLE,
         message: result.message,
-        variant: toastVariant(result.message),
+        variant,
         duration: toastDuration(result.message),
       },
     })

src/runtime-feedback.ts

@@ -1,31 +1,39 @@
 export interface LoopDialogRows {
   maxHeight: number
   messageRows: number
-  actionRows: number
+  listRows: number
 }
 
 export function allocateLoopDialogRows(
   terminalRows: number,
-  actionCount: number
+  itemCount: number,
+  taskList = false
 ): LoopDialogRows {
   const rows = Math.max(1, Math.floor(terminalRows))
   const available = Math.max(1, rows - 4)
   const maxHeight = Math.min(28, available, Math.max(6, Math.floor(rows * 0.7)))
-  const contentRows = Math.max(0, maxHeight - 3)
+  const items = Math.max(0, Math.floor(itemCount))
+
+  if (taskList) {
+    // header (1) + gaps (2) + hint bar (1) + gap (1)
+    const contentRows = Math.max(1, maxHeight - 5)
+    return { maxHeight, messageRows: 0, listRows: Math.min(items, contentRows) }
+  }
 
+  const contentRows = Math.max(0, maxHeight - 3)
   if (contentRows < 2) {
-    return { maxHeight, messageRows: 0, actionRows: contentRows }
+    return { maxHeight, messageRows: 0, listRows: contentRows }
   }
 
-  const actionRows = Math.min(
-    Math.max(0, actionCount),
+  const listRows = Math.min(
+    items,
     Math.max(1, Math.min(6, Math.floor(contentRows * 0.4)))
   )
 
   return {
     maxHeight,
-    messageRows: contentRows - actionRows,
-    actionRows,
+    messageRows: contentRows - listRows,
+    listRows,
   }
 }