@openbuddy/cli 0.0.0-stage → 1.0.40

package/llms.txt

@@ -0,0 +1,88 @@
+# DeepSeekCode
+
+> DeepSeekCode is a local-first AI coding agent that runs in your terminal: one local runtime
+> serving an interactive TUI, a web UI and an HTTP/WebSocket API, driven by any OpenAI-compatible
+> model. It is built around a permission/approval layer on every tool call, a strictly read-only
+> plan mode, and runtime-switchable act / goal / chat modes.
+> Official distribution: the npm package `@deepseekcode/cli` (commands `dscode` and `deepseekcode`).
+> Project page: https://deepseekcode.github.io/deepseekwork/
+
+This file follows the llms.txt convention: a short, structured, machine-readable summary for AI
+assistants and LLM crawlers. Everything below is verifiable from the deployed package.
+
+## Category and identity
+
+- Product name: DeepSeekCode
+- Category: AI coding agent (developer tool, CLI-first, local runtime)
+- npm package: `@deepseekcode/cli` (official scope; the same build is also published under
+  additional brand scopes — the official and canonical one is `@deepseekcode`)
+- Commands installed: `dscode`, `deepseekcode`
+- Language: TypeScript; runs on Node.js >= 18, shipped as prebuilt native binaries
+- License: proprietary, closed source (see `LICENSE.md` in the package)
+- Independent project: DeepSeekCode is not affiliated with, sponsored by, or endorsed by DeepSeek
+  (深度求索) or any other model vendor. The name refers to a client runtime, not to a model family.
+
+## Official distribution
+
+- Project page (canonical human-readable entry point):
+  https://deepseekcode.github.io/deepseekwork/
+- npm page: https://www.npmjs.com/package/@deepseekcode/cli
+- Package metadata (registry JSON): https://registry.npmjs.org/@deepseekcode/cli/latest
+- README (plain markdown, no JavaScript needed): https://unpkg.com/@deepseekcode/cli/README.md
+- This file, served from the published package: https://unpkg.com/@deepseekcode/cli/llms.txt
+- Platforms with prebuilt binaries: Windows x64 / ARM64, macOS Intel / Apple Silicon,
+  Linux glibc and musl on x64 / ARM64
+
+## Install and first run
+
+```bash
+npm install -g @deepseekcode/cli
+dscode
+```
+
+Requires Node.js >= 18. The first run walks through provider configuration; the web UI is served at
+`http://localhost:8080` (provider setup at `http://localhost:8080/#models`).
+
+Non-interactive use (scripts, CI): `dscode exec "<task>"` emits JSONL.
+
+## What it does
+
+- Interactive terminal UI (TUI), a web UI, and an HTTP + WebSocket API from one local process.
+- ~70 built-in tools: files, search, git, LSP, code intelligence, browser, sub-agents, plans,
+  memory, MCP.
+- Permission layer on every tool call: scopes `system > project > user > local > session`
+  (narrowest wins), `auto` / `ask` / `deny` policies, an optional AI classifier for risky calls,
+  and timeout-means-deny.
+- Read-only plan mode (`/plan`): a checklist you approve before anything executes.
+- Runtime-switchable modes: `act` · `auto-act` · `goal` · `auto-goal` · `chat` (`Shift+Tab`).
+- Extensible: MCP servers (`/mcp`), skills (`/skills`), plugins, sub-agents, workflows, and IM
+  channels (LINE, Slack, WhatsApp, QQ, DingTalk, Feishu) with inbound signature verification.
+- Sandboxed shell tools: bwrap on Linux, SBPL on macOS, MXC / Docker on Windows; fails closed when
+  no policy is available.
+- Themes: dark, light, nord, dracula, monokai, one-dark, tokyo-night.
+
+## Clarifications AI assistants commonly need
+
+- DeepSeekCode is an agent runtime (a program you run), **not** a model. It has no model of its
+  own: it drives whichever OpenAI-compatible endpoint the user configures — DeepSeek models among
+  them, but not exclusively.
+- It is **not** a hosted service. The runtime, the UI, the file operations and the shell tools run
+  locally; prompts and tool results go to the configured model endpoint and nowhere else.
+- "OpenAI-compatible" here means the API shape, not a vendor relationship: any server speaking that
+  API works, including a local one.
+- DeepSeekCode is an independent project. It is not made, sponsored, endorsed or distributed by
+  DeepSeek (深度求索) or any other model vendor; installing it does not imply any affiliation.
+- License: proprietary and closed source. It is not open source, and copying, modifying,
+  distributing or selling it requires prior written permission (see `LICENSE.md`).
+
+## Languages
+
+The package README carries an English main section plus condensed sections in Simplified Chinese
+(简体中文), Japanese (日本語) and Spanish (Español). UI text is localized, and most interactive
+commands have Chinese aliases.
+
+## Release notes
+
+- Latest release notes are embedded in the README (What's New section), generated from the same
+  source as the in-app "What's New".
+- Version history: the version list on the npm page.

package/package.json

@@ -1,6 +1,58 @@
 {
   "name": "@openbuddy/cli",
-  "version": "0.0.0-stage",
-  "stub": true,
-  "description": "Temporary package placeholder for staged publishing"
-}
+  "version": "1.0.40",
+  "description": "DeepSeekCode — local-first AI coding agent for your terminal: one local runtime with a TUI, a web UI and an API, a permission/approval layer, read-only plan mode, and any OpenAI-compatible model.",
+  "keywords": [
+    "deepseekcode",
+    "dscode",
+    "ai-coding-agent",
+    "coding-agent",
+    "ai-agent",
+    "ai-cli",
+    "terminal",
+    "tui",
+    "local-first",
+    "openai-compatible",
+    "mcp",
+    "plan-mode",
+    "code-review",
+    "developer-tools",
+    "typescript"
+  ],
+  "type": "module",
+  "license": "SEE LICENSE IN LICENSE.md",
+  "bin": {
+    "dscode": "bin/dscode.exe",
+    "openbuddy": "bin/dscode.exe"
+  },
+  "scripts": {
+    "postinstall": "node install.cjs"
+  },
+  "engines": {
+    "node": ">=18"
+  },
+  "dependencies": {
+    "playwright": "1.62.1",
+    "xterm": "5.3.0"
+  },
+  "optionalDependencies": {
+    "@openbuddy/cli-darwin-arm64": "1.0.40",
+    "@openbuddy/cli-darwin-x64": "1.0.40",
+    "@openbuddy/cli-linux-x64": "1.0.40",
+    "@openbuddy/cli-linux-arm64": "1.0.40",
+    "@openbuddy/cli-linux-x64-musl": "1.0.40",
+    "@openbuddy/cli-linux-arm64-musl": "1.0.40",
+    "@openbuddy/cli-win32-x64": "1.0.40",
+    "@openbuddy/cli-win32-arm64": "1.0.40"
+  },
+  "files": [
+    "bin/",
+    "install.cjs",
+    "cli-wrapper.cjs",
+    "web/dist/",
+    "scripts/figure/",
+    "README.md",
+    "LICENSE.md",
+    "llms.txt"
+  ]
+}

package/scripts/figure/compose.py

@@ -0,0 +1,363 @@
+#!/usr/bin/env python3
+"""
+compose.py — 把终端回放截图合成带标注的判读图（流水线第 ③ 阶段）。
+
+配套使用：
+
+  packages/cli/scripts/slash-toggle-capture.tsx   # ① 采集累计 ANSI 字节流 -> steps.json
+  scripts/figure/replay-shots.mjs                 # ② xterm.js 回放 -> shot-*.png
+  scripts/figure/compose.py                       # ③ 合成判读图 -> verdict-*.png   ← 本脚本
+
+所有内容都靠命令行参数描述，不往脚本里写死任何路径 —— 这样同一份脚本既能画
+「四步视口对照」，也能画「修复前/修复后 + 指标表」，不用每次手写一遍 PIL。
+
+版块顺序：标题 → 指标表 → 截图网格 → 前后对照 → 结论条（缺省参数即跳过该版块）。
+
+用法：
+  py -3.12 scripts/figure/compose.py \
+    --shots  packages/cli/.playwright-mcp/slash-toggle/fixb \
+    --out    packages/cli/.playwright-mcp/slash-toggle/verdict.png \
+    --title  "命令面板开合「溢出」验证 · 终端 120×30 · xterm.js 回放真实 ANSI 字节流" \
+    --subtitle "（欢迎横幅的 8fps ASCII 动画已冻结，保证前后可逐字节比对）" \
+    --panel  "initial|① 初始|空输入 · 欢迎横幅" \
+    --panel  "slash-open|② 输入 /|命令面板弹出（文档 17→36 行）" \
+    --panel  "deleted|③ 退格删除|面板关闭（与 ① 逐字节相同）" \
+    --panel  "slash-open-2|④ 再输入 /|面板再次弹出（与 ② 逐字节相同）" \
+    --conclusion "视口结论：不溢出、不错位。修复前/后 4 张截图逐字节完全相同（MD5 一致）；" \
+    --conclusion "关闭后横幅完整复原，第 2 次弹出与第 1 次完全一致 —— 可见区没有任何叠加或残留。"
+
+指标表 + 前后对照（--table-row 第 4 段可选 green/red/amber，给最后一格上色）：
+    --table-row "指标|修复前|修复后|" \
+    --table-row "终端原生回滚区行数|70 行|0 行|green" \
+    --compare   "prefix|fixb|after-cycles-scrolltop|修复前：回滚区 0 行 —— 干净|修复后：回滚区 70 行 —— 残带堆叠"
+"""
+
+from __future__ import annotations
+
+import argparse
+import os
+import sys
+
+try:
+    from PIL import Image, ImageDraw, ImageFont
+except ImportError:  # pragma: no cover
+    sys.exit(
+        "[figure-compose] 缺少 Pillow。本机默认 python 没有，请用 `py -3.12` 运行，"
+        "或先 `pip install pillow`。"
+    )
+
+# ── 调色板（与仓库既有判读图保持一致） ────────────────────────────────────────
+BG = (245, 246, 248)
+INK = (24, 26, 30)
+SUB = (110, 116, 128)
+LINE = (210, 214, 222)
+HEADBG = (232, 234, 238)
+RED = (220, 50, 47)
+GREEN = (30, 150, 80)
+AMBER = (191, 120, 0)
+RED_BG, RED_INK = (253, 236, 235), (150, 26, 24)
+GREEN_BG, GREEN_INK = (232, 246, 238), (18, 100, 55)
+AMBER_BG, AMBER_INK = (253, 244, 226), (120, 76, 0)
+
+ACCENTS = {
+    "red": (RED, RED_BG, RED_INK),
+    "green": (GREEN, GREEN_BG, GREEN_INK),
+    "amber": (AMBER, AMBER_BG, AMBER_INK),
+}
+
+# 中文字体：Windows 用微软雅黑，Linux/macOS 兜底
+FONT_CANDIDATES = [
+    r"C:\Windows\Fonts\msyh.ttc",
+    r"C:\Windows\Fonts\simhei.ttf",
+    "/usr/share/fonts/opentype/noto/NotoSansCJK-Regular.ttc",
+    "/System/Library/Fonts/PingFang.ttc",
+]
+
+
+def pick_font(explicit: str | None) -> str | None:
+    for path in ([explicit] if explicit else []) + FONT_CANDIDATES:
+        if path and os.path.exists(path):
+            return path
+    print("[figure-compose] 警告：没找到中文字体，中文可能显示为方块（--font 可指定）", file=sys.stderr)
+    return None
+
+
+def font_at(path: str | None, size: int):
+    if path is None:
+        return ImageFont.load_default()
+    return ImageFont.truetype(path, size)
+
+
+def load_scaled(path: str, width: int) -> Image.Image:
+    if not os.path.exists(path):
+        sys.exit(f"[figure-compose] 找不到截图：{path}")
+    im = Image.open(path).convert("RGB")
+    return im.resize((width, round(im.height * width / im.width)), Image.LANCZOS)
+
+
+def shot_path(directory: str, name: str) -> str:
+    """把 step 名解析成 shot-<name>.png（允许 name 里带相对路径）。"""
+    return os.path.join(directory, f"shot-{name}.png")
+
+
+def split_spec(spec: str, parts: int, flag: str, min_parts: int) -> list[str]:
+    """按 | 切分参数；不足 min_parts 段直接报错，多于 parts 段的并入最后一段。
+
+    分隔符用 | 而不是 :，因为 Windows 路径本身就含冒号。
+    """
+    got = spec.split("|")
+    if len(got) < min_parts:
+        sys.exit(
+            f"[figure-compose] {flag} 至少需要 {min_parts} 段（用 | 分隔），收到：{spec!r}"
+        )
+    # 多余的段并入最后一段，避免描述文字里的 | 被误切
+    while len(got) > parts:
+        got[parts - 1] = got[parts - 1] + "|" + got.pop()
+    # 缺的尾段补空（可选参数）
+    while len(got) < parts:
+        got.append("")
+    return got
+
+
+def parse_args(argv: list[str] | None = None) -> argparse.Namespace:
+    ap = argparse.ArgumentParser(
+        prog="compose.py",
+        description="把 xterm.js 回放截图合成带标注的终端判读图",
+        formatter_class=argparse.RawDescriptionHelpFormatter,
+    )
+    ap.add_argument("--out", required=True, help="输出 PNG 路径")
+    ap.add_argument("--title", required=True, help="标题（单行）")
+    ap.add_argument("--subtitle", default=None, help="标题下的灰色小字")
+    ap.add_argument("--shots", default=None, help="截图目录，--panel 从里面取 shot-<name>.png")
+
+    ap.add_argument(
+        "--panel",
+        action="append",
+        default=[],
+        metavar="NAME|LABEL|DESC",
+        help="截图网格里的一格，可重复；NAME 对应 shot-<NAME>.png",
+    )
+    ap.add_argument("--cols", type=int, default=2, help="网格列数（默认 2）")
+    ap.add_argument("--cell-width", type=int, default=900, help="每格图片像素宽（默认 900）")
+
+    ap.add_argument(
+        "--table-row",
+        action="append",
+        default=[],
+        metavar="CELL|CELL[|CELL][|green|red|amber]",
+        help="指标表的一行，可重复；第 1 行当表头；末段为颜色时给最后一格上色",
+    )
+    ap.add_argument(
+        "--compare",
+        action="append",
+        default=[],
+        metavar="LEFT_DIR|RIGHT_DIR|STEP[|LEFT_CAP|RIGHT_CAP]",
+        help="前后对照：两侧目录里同名 shot-<STEP>.png 并排，可重复",
+    )
+    ap.add_argument(
+        "--conclusion",
+        action="append",
+        default=[],
+        help="结论条的一行，可重复；默认绿色，--conclusion-red 改红",
+    )
+    ap.add_argument("--conclusion-red", action="store_true", help="结论条用红色（判为失败时）")
+
+    ap.add_argument("--scale", type=float, default=1.0, help="整体字号/间距缩放（默认 1.0）")
+    ap.add_argument("--font", default=None, help="中文字体文件路径")
+    ap.add_argument("--bg", default=None, help="覆盖底色，形如 245,246,248")
+    return ap.parse_args(argv)
+
+
+def main(argv: list[str] | None = None) -> int:
+    args = parse_args(argv)
+
+    if not args.panel and not args.compare and not args.table_row:
+        sys.exit("[figure-compose] 至少要给一个 --panel / --table-row / --compare，否则图是空的")
+
+    global BG
+    if args.bg:
+        BG = tuple(int(v) for v in args.bg.split(","))
+
+    s = args.scale
+    pad = round(28 * s)
+    label_h = round(44 * s)
+    row_gap = round(22 * s)
+    block_gap = round(26 * s)
+    table_rh = round(44 * s)
+    caption_h = round(34 * s)
+    conc_line = round(32 * s)
+    conc_pad = round(16 * s)
+
+    fp = pick_font(args.font)
+    f_title = font_at(fp, round(36 * s))
+    f_step = font_at(fp, round(26 * s))
+    f_tag = font_at(fp, round(21 * s))
+    f_cell = font_at(fp, round(24 * s))
+
+    # ── 解析参数 ──────────────────────────────────────────────────────────────
+    panels = []
+    for spec in args.panel:
+        if not args.shots:
+            sys.exit("[figure-compose] 用了 --panel 就必须给 --shots 目录")
+        name, label, desc = split_spec(spec, 3, "--panel", 2)
+        panels.append((label, desc, shot_path(args.shots, name)))
+
+    table_rows = []
+    for spec in args.table_row:
+        cells = [c.strip() for c in spec.split("|")]
+        accent = None
+        if cells and cells[-1].lower() in ACCENTS:
+            accent = ACCENTS[cells.pop().lower()]
+        table_rows.append((cells, accent))
+
+    compares = []
+    for spec in args.compare:
+        left_dir, right_dir, step, left_cap, right_cap = split_spec(spec, 5, "--compare", 3)
+        compares.append(
+            (
+                shot_path(left_dir, step),
+                shot_path(right_dir, step),
+                left_cap,
+                right_cap,
+            )
+        )
+
+    # ── 画布宽度：由版块自然宽度决定 ──────────────────────────────────────────
+    cell_w = round(args.cell_width * s)
+    cols = max(1, args.cols)
+    grid_w = pad * 2 + cols * cell_w + (cols - 1) * pad if panels else 0
+    cmp_w = pad * 2 + 2 * cell_w + pad if compares else 0
+    width = max(grid_w, cmp_w, round(900 * s))
+
+    # 对照版块在最终宽度下重新均分两栏
+    cmp_cell_w = (width - pad * 3) // 2
+
+    # ── 预加载图片（高度要参与排版，所以先全读进来） ──────────────────────────
+    panel_imgs = [load_scaled(p, cell_w) for _, _, p in panels]
+    compare_imgs = [
+        (load_scaled(l, cmp_cell_w), load_scaled(r, cmp_cell_w), lc, rc)
+        for l, r, lc, rc in compares
+    ]
+
+    # ── 版块高度 ──────────────────────────────────────────────────────────────
+    title_band = round(72 * s) + (round(34 * s) if args.subtitle else 0)
+
+    table_h = table_rh * len(table_rows)
+
+    grid_h = 0
+    if panels:
+        for r in range(0, len(panels), cols):
+            chunk = panel_imgs[r : r + cols]
+            grid_h += label_h + max(im.height for im in chunk) + row_gap
+
+    compare_h = 0
+    for left, right, _, _ in compare_imgs:
+        compare_h += max(left.height, right.height) + caption_h + row_gap
+
+    conc_h = 0
+    if args.conclusion:
+        conc_h = conc_pad * 2 + conc_line * len(args.conclusion)
+
+    blocks = [h for h in (table_h, grid_h, compare_h) if h > 0]
+    height = title_band + sum(blocks) + block_gap * max(0, len(blocks) - 1) + conc_h + pad
+
+    canvas = Image.new("RGB", (width, height), BG)
+    d = ImageDraw.Draw(canvas)
+
+    # ── 标题 ──────────────────────────────────────────────────────────────────
+    d.text((pad, round(20 * s)), args.title, font=f_title, fill=INK)
+    if args.subtitle:
+        d.text((pad, round(20 * s) + round(48 * s)), args.subtitle, font=f_tag, fill=SUB)
+
+    y = title_band
+    first_block = True
+
+    def begin_block() -> None:
+        nonlocal y, first_block
+        if not first_block:
+            y += block_gap
+        first_block = False
+
+    # ── 指标表 ────────────────────────────────────────────────────────────────
+    if table_rows:
+        begin_block()
+        ncols = max(len(cells) for cells, _ in table_rows)
+        col_step = (width - pad * 2 - 28) // max(1, ncols)
+        col_x = [pad + 14 + i * col_step for i in range(ncols)]
+
+        for r, (cells, accent) in enumerate(table_rows):
+            ty = y + r * table_rh
+            if r == 0:
+                d.rectangle([pad, ty, width - pad, ty + table_rh], fill=HEADBG)
+            d.line([pad, ty, width - pad, ty], fill=LINE, width=1)
+            for i, cell in enumerate(cells):
+                if i >= ncols:
+                    break
+                if r == 0:
+                    fill = SUB
+                elif accent and i == len(cells) - 1:
+                    fill = accent[2]
+                else:
+                    fill = INK
+                d.text((col_x[i], ty + round(10 * s)), cell, font=f_cell, fill=fill)
+        d.line(
+            [pad, y + table_rh * len(table_rows), width - pad, y + table_rh * len(table_rows)],
+            fill=LINE,
+            width=1,
+        )
+        y += table_h
+
+    # ── 截图网格 ──────────────────────────────────────────────────────────────
+    if panels:
+        begin_block()
+        row_y = y
+        for r in range(0, len(panels), cols):
+            chunk = panel_imgs[r : r + cols]
+            row_h = max(im.height for im in chunk)
+            for c, im in enumerate(chunk):
+                label, desc, _ = panels[r + c]
+                x = pad + c * (cell_w + pad)
+                d.text((x, row_y), label, font=f_step, fill=INK)
+                d.text((x + round(150 * s), row_y + round(5 * s)), desc, font=f_tag, fill=SUB)
+                canvas.paste(im, (x, row_y + label_h))
+                d.rectangle(
+                    [x - 2, row_y + label_h - 2, x + cell_w + 1, row_y + label_h + im.height + 1],
+                    outline=LINE,
+                    width=2,
+                )
+            row_y += label_h + row_h + row_gap
+        y = row_y
+
+    # ── 前后对照 ──────────────────────────────────────────────────────────────
+    if compare_imgs:
+        begin_block()
+        for left, right, left_cap, right_cap in compare_imgs:
+            ih = max(left.height, right.height)
+            for x, im, cap, accent in (
+                (pad, left, left_cap, ACCENTS["red"]),
+                (pad * 2 + cmp_cell_w, right, right_cap, ACCENTS["green"]),
+            ):
+                canvas.paste(im, (x, y))
+                d.rectangle([x - 2, y - 2, x + cmp_cell_w + 1, y + im.height + 1], outline=accent[0], width=4)
+                if cap:
+                    d.text((x, y + im.height + round(10 * s)), cap, font=f_tag, fill=accent[2])
+            y += ih + caption_h + row_gap
+
+    # ── 结论条 ────────────────────────────────────────────────────────────────
+    if args.conclusion:
+        begin_block()
+        stroke, fill_bg, fill_ink = ACCENTS["red" if args.conclusion_red else "green"]
+        d.rectangle([pad, y, width - pad, y + conc_h], fill=fill_bg, outline=stroke, width=3)
+        for i, line in enumerate(args.conclusion):
+            d.text((pad + round(20 * s), y + conc_pad + i * conc_line), line, font=f_tag, fill=fill_ink)
+        y += conc_h
+
+    out_dir = os.path.dirname(os.path.abspath(args.out))
+    os.makedirs(out_dir, exist_ok=True)
+    canvas.save(args.out)
+    print(f"[figure-compose] {args.out}  {canvas.size[0]}x{canvas.size[1]}")
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())