@remixmate/cli 0.1.0

package/skills/video-parser/scripts/parse_via_render.py

@@ -0,0 +1,148 @@
+#!/usr/bin/env python3
+"""
+视频解构脚本 — 通过 ab-render HTTP 接口实现，无需本地 ffmpeg。
+
+调用链：
+  POST $RENDER_API_URL/parse        → { taskId }
+  POST $RENDER_API_URL/parseStatus  → { status, progress, result?, error? }
+
+必要环境变量（由 ab-agent 自动注入）：
+  RENDER_API_URL   ab-render 服务地址，例如 http://localhost:3000
+  PRIV_TOKEN       ab-api 私有 token
+
+可选环境变量：
+  CONVERSATION_ID  对话 ID，用于文件关联
+"""
+
+import argparse
+import json
+import os
+import sys
+import time
+import urllib.error
+import urllib.request
+
+# ─── 环境变量 ─────────────────────────────────────────────────────────────────
+
+RENDER_API_URL = os.environ.get("RENDER_API_URL", "").rstrip("/")
+PRIV_TOKEN = os.environ.get("PRIV_TOKEN", "")
+CONVERSATION_ID = os.environ.get("CONVERSATION_ID", "")
+
+# ─── HTTP 工具 ────────────────────────────────────────────────────────────────
+
+def api_post(path: str, body: dict, timeout: int = 30) -> dict:
+    """向 ab-render 发送 POST 请求，返回解析后的 JSON。"""
+    url = f"{RENDER_API_URL}{path}"
+    data = json.dumps(body).encode()
+
+    headers: dict[str, str] = {
+        "Content-Type": "application/json",
+        "X-Priv-Token": PRIV_TOKEN,
+    }
+    if CONVERSATION_ID:
+        headers["x-conversation-id"] = CONVERSATION_ID
+
+    req = urllib.request.Request(url, data=data, headers=headers, method="POST")
+    try:
+        with urllib.request.urlopen(req, timeout=timeout) as resp:
+            return json.loads(resp.read())
+    except urllib.error.HTTPError as e:
+        body_text = e.read().decode(errors="replace")
+        raise RuntimeError(f"HTTP {e.code}: {body_text}") from e
+
+
+def log(msg: str, silent: bool = False) -> None:
+    if not silent:
+        print(msg, flush=True)
+
+
+# ─── 主逻辑 ───────────────────────────────────────────────────────────────────
+
+def main() -> None:
+    parser = argparse.ArgumentParser(
+        description="Video deconstruction — calls the ab-render /parse endpoint (no local ffmpeg dependency)"
+    )
+    parser.add_argument("--url", required=True, help="Remote video URL")
+    parser.add_argument(
+        "--scene-threshold", type=float, default=0.3,
+        help="Scene-change sensitivity 0.0-1.0 (default 0.3)",
+    )
+    parser.add_argument("--skip-asr", action="store_true", help="Skip the ASR step")
+    parser.add_argument("--skip-keyframes", action="store_true", help="Skip keyframe extraction")
+    parser.add_argument(
+        "--json-output", action="store_true",
+        help="Pipeline mode: emit only the final JSON to stdout",
+    )
+    args = parser.parse_args()
+
+    # ── Pre-flight check ─────────────────────────────────────────────────────
+    if not RENDER_API_URL:
+        print(
+            "❌ RENDER_API_URL env var is not configured.\n"
+            "   Add render_api_url: \"http://<ab-render-host>:<port>\" to Nacos ab-agent.yaml",
+            file=sys.stderr,
+        )
+        sys.exit(1)
+
+    # ── 1. Submit the deconstruction task ───────────────────────────────────
+    log(f"[video-parser] Submitting deconstruction task: {args.url[:80]}", args.json_output)
+
+    resp = api_post("/parse", {
+        "videoUrl": args.url,
+        "sceneThreshold": args.scene_threshold,
+        "skipAsr": args.skip_asr,
+        "skipKeyframes": args.skip_keyframes,
+    })
+
+    if resp.get("code") != 0:
+        print(f"❌ submission failed: {resp.get('msg', 'unknown error')}", file=sys.stderr)
+        sys.exit(1)
+
+    task_id: str = resp["data"]["taskId"]
+    log(f"[video-parser] task accepted, taskId={task_id}", args.json_output)
+
+    # ── 2. Poll status until completion ─────────────────────────────────────
+    max_wait_sec = 600   # Wait up to 10 minutes
+    poll_interval = 3    # Poll every 3 seconds
+    elapsed = 0
+
+    while elapsed < max_wait_sec:
+        time.sleep(poll_interval)
+        elapsed += poll_interval
+
+        try:
+            status_resp = api_post("/parseStatus", {"taskId": task_id})
+        except Exception as e:  # noqa: BLE001
+            log(f"[video-parser] polling error (will retry): {e}", args.json_output)
+            continue
+
+        if status_resp.get("code") != 0:
+            log(f"[video-parser] polling returned an error code (will retry): {status_resp.get('msg')}", args.json_output)
+            continue
+
+        task = status_resp["data"]
+        status = task.get("status", "unknown")
+        progress = task.get("progress", 0)
+
+        log(f"[video-parser] status={status}  progress={int(progress * 100)}%  elapsed={elapsed}s", args.json_output)
+
+        if status == "succeeded":
+            result = task.get("result")
+            if args.json_output:
+                print(json.dumps(result, ensure_ascii=False, indent=2))
+            else:
+                print("\n✅ Deconstruction complete")
+                print(json.dumps(result, ensure_ascii=False, indent=2))
+            return
+
+        if status == "failed":
+            error = task.get("error", "unknown error")
+            print(f"❌ deconstruction failed: {error}", file=sys.stderr)
+            sys.exit(1)
+
+    print(f"❌ deconstruction timed out (over {max_wait_sec}s)", file=sys.stderr)
+    sys.exit(1)
+
+
+if __name__ == "__main__":
+    main()

package/skills/video-parser/skill.json

@@ -0,0 +1,21 @@
+{
+  "name": "video-parser",
+  "toolName": "video_parser",
+  "tier": "tool",
+  "title": "Video Deconstruction",
+  "description": "Video deconstruction: split a video into reusable assets (audio extraction, ASR transcript, keyframe extraction, scene segmentation).",
+  "envVars": ["PRIV_TOKEN", "MM_API_BASE_URL", "AGENT_NAME", "RENDER_API_URL"],
+  "scriptPath": "scripts/parse_via_render.py",
+  "parameters": {
+    "type": "object",
+    "properties": {
+      "url": { "type": "string", "description": "Remote video URL (or use local)" },
+      "local": { "type": "string", "description": "Local video path (or use url)" },
+      "scene_threshold": { "type": "number", "description": "Scene-cut detection threshold 0.0-1.0 (default 0.3)" },
+      "skip_asr": { "type": "boolean", "description": "Skip the ASR step" },
+      "skip_keyframes": { "type": "boolean", "description": "Skip the keyframe-extraction step" },
+      "json_output": { "type": "boolean", "description": "Pipeline mode — emit JSON only on stdout" }
+    },
+    "required": []
+  }
+}

package/skills/video-parser/version.json

@@ -0,0 +1,7 @@
+{
+  "skillName": "video-parser",
+  "repoName": "agent-skill-media-maker",
+  "skillId": "553",
+  "version": "V1",
+  "skillDescription": "视频解构技能，将视频拆解为可复用的内容资产（音频、ASR 文本、关键帧、场景分段）。\n\n当用户提到以下任何需求时，立即使用本 skill：\n- 视频解构、视频拆解、视频分析、解析视频\n- 提取关键帧、提取视频文案、视频转文字\n- 视频内容资产、视频素材提取\n\n即使用户没有明确说「解构」，只要他们想要从视频中提取文案、关键帧或结构化信息，也要使用本 skill。"
+}

package/skills/web-capture/SKILL.md

@@ -0,0 +1,183 @@
+---
+name: web-capture
+description: |
+  Web page capture skill. Drives a headless browser (Playwright) to turn a URL into reusable visual assets — viewport / full-page screenshots, or a smooth top-to-bottom scrolling video (MP4). These assets are the upstream material for the `screen-walkthrough` template.
+
+  Use this skill as soon as the user mentions any of these intents:
+  - Screenshot a web page / website / a specific URL, full-page screenshot
+  - Record a page scrolling top-to-bottom into a video, "scroll video" of a site
+  - Produce screen-recording / walkthrough footage of a GitHub repo, landing page, or product page
+
+  Even when the user does not say "Playwright" or "browser", use this skill whenever they want to turn a web address into a screenshot or a scrolling demo video.
+triggers:
+  - Screenshot a web page / website / a specific URL, full-page screenshot
+  - Record a page scrolling top-to-bottom into a video, "scroll video" of a site
+  - Produce screen-recording / walkthrough footage of a GitHub repo, landing page, or product page
+---
+
+# Web Page Capture Skill
+
+Drives a **headless browser via Playwright (Python)** to capture a web page as reusable visual assets:
+
+- **Screenshots** — viewport, full scrollable page, or a single element.
+- **Scroll video** — opens the page and smoothly scrolls top-to-bottom, capturing frames and assembling a deterministic MP4 with ffmpeg.
+- **Record video** — Playwright's native context recording (webm), transcoded to MP4.
+
+The MP4 / PNG output is designed to feed the **`screen-walkthrough`** template's screen stage.
+
+## Dependencies
+
+Aligned with method 1 of the design — the runtime provides the browser, the skill does not bundle it.
+
+- **Python 3** with **Playwright**: `pip install playwright` then `playwright install chromium` (one-time; downloads the browser binary).
+- **ffmpeg**: must be installed locally (`brew install ffmpeg`) — used to assemble the scroll-video frames and transcode native recordings to MP4. Same dependency the `video-parser` skill relies on.
+
+The script checks for both at startup and prints an actionable install hint if either is missing (`__progress__` phase `web-capture:preflight`).
+
+## Multi-user isolation & security
+
+ab-agent spawns one fresh `python3` process per tool call (`spawn('python3', ...)`), and this script launches its own Playwright browser instance per run and closes it on exit. So **concurrent users are isolated at the process + browser-instance level** — no cookie/session bleed between users. The skill hardens a few defaults on top of that:
+
+- **Unique output dir**: when `--output-dir` is omitted, output goes to a unique system temp dir (`tempfile.mkdtemp`), never the script's cwd. Concurrent calls can't collide or pollute `skills/`.
+- **Temp cleanup**: per-frame images (`_frames/`) and native-recording webm (`_rec/`) are removed in a `finally` block, so failures don't leak temp data.
+- **SSRF guard**: the user-supplied `url` is validated before any browser launch — only `http`/`https`, and the resolved IP must not be private / loopback / link-local / reserved / cloud-metadata (`169.254.169.254`). Rejection exits with code 5.
+- **Sandbox on by default**: Chromium keeps its sandbox; `--no-sandbox` is only added when `WEB_CAPTURE_NO_SANDBOX=1`.
+
+Not handled in-skill (orchestration concern): **concurrency limits**. Each browser is ~150–300MB; many simultaneous `scroll-video` calls can exhaust host memory. Gate concurrency at the ab-agent layer or via host resources.
+
+### Environment variables
+
+| Env var | Description | Default |
+|---------|-------------|---------|
+| `WEB_CAPTURE_BROWSER` | Default browser engine (`--browser` overrides). | `chromium` |
+| `WEB_CAPTURE_OUTPUT_BASE` | Base dir for the auto-created temp output dir. | system tempdir |
+| `WEB_CAPTURE_NO_SANDBOX` | Set `1` to add `--no-sandbox` (only for containers that can't use user namespaces). | unset (sandbox on) |
+| `WEB_CAPTURE_ALLOW_PRIVATE_HOSTS` | Set `1` to disable the private/internal-IP SSRF block (single-tenant / trusted intranet only). | unset (block on) |
+| `PLAYWRIGHT_BROWSERS_PATH` | Playwright browser binary location (provided by the runtime). | Playwright default |
+
+## Operations
+
+1. **Script path**: read the system-injected `Base directory for this skill: <path>` as `<SkillDir>` and always use `<SkillDir>/scripts/web_capture.py`. Never hard-code paths.
+2. **Pick the mode**: `screenshot` (default), `scroll-video`, or `record-video`.
+3. **Run the capture** and wait for it to finish.
+4. **Inspect the result**: a `capture.json` manifest plus the produced `*.png` / `*.mp4` under `--output-dir`.
+
+### Viewport screenshot
+
+```bash
+python3 <SkillDir>/scripts/web_capture.py \
+  --url "https://github.com/microsoft/playwright" \
+  --mode screenshot
+```
+
+### Full-page screenshot (whole scrollable page)
+
+```bash
+python3 <SkillDir>/scripts/web_capture.py \
+  --url "https://example.com" \
+  --mode screenshot \
+  --full-page
+```
+
+### Element-only screenshot
+
+```bash
+python3 <SkillDir>/scripts/web_capture.py \
+  --url "https://example.com" \
+  --mode screenshot \
+  --selector "#readme"
+```
+
+### Scrolling video (top-to-bottom)
+
+```bash
+python3 <SkillDir>/scripts/web_capture.py \
+  --url "https://github.com/microsoft/playwright" \
+  --mode scroll-video \
+  --width 1280 --height 720 \
+  --fps 30 \
+  --scroll-duration 8
+```
+
+Produces a deterministic `scroll.mp4` — fps, resolution and scroll speed are all controllable, which keeps the output aligned with the `screen-walkthrough` timeline.
+
+### Native recording (webm → mp4)
+
+```bash
+python3 <SkillDir>/scripts/web_capture.py \
+  --url "https://example.com" \
+  --mode record-video \
+  --scroll-duration 6
+```
+
+Simpler than frame-capture but the frame rate is not precisely controllable; prefer `scroll-video` when timing matters.
+
+### JSON pipeline mode
+
+```bash
+python3 <SkillDir>/scripts/web_capture.py \
+  --url "https://example.com" \
+  --mode screenshot \
+  --json-output
+```
+
+Prints only the final `capture.json` payload to stdout — no progress lines — suitable for piping into downstream skills.
+
+See every option: `python3 <SkillDir>/scripts/web_capture.py --help`.
+
+## Common CLI flags
+
+| Flag | Description | Default |
+|------|-------------|---------|
+| `--url` | Target page URL (http/https). | required |
+| `--mode` | `screenshot` / `scroll-video` / `record-video`. | `screenshot` |
+| `-o` / `--output-dir` | Output directory. | unique system temp dir |
+| `--width` | Viewport width (px). | `1280` |
+| `--height` | Viewport height (px). | `720` |
+| `--device-scale-factor` | Device pixel ratio (crisp / retina). | `2` |
+| `--full-page` | Screenshot mode: capture the whole scrollable page. | off |
+| `--selector` | Screenshot mode: capture only this CSS element. | none |
+| `--fps` | Scroll / record video frame rate. | `30` |
+| `--scroll-duration` | Seconds for the full top-to-bottom scroll. | `8` |
+| `--browser` | `chromium` / `firefox` / `webkit`. | `chromium` |
+| `--wait-until` | `load` / `domcontentloaded` / `networkidle`. | `networkidle` |
+| `--wait-ms` | Extra wait after load before capture (ms). | `1500` |
+| `--json-output` | Pipeline mode — JSON-only stdout. | off |
+
+## Output layout
+
+```
+<output-dir>/
+├── capture.json          # asset manifest (mode, url, viewport, produced files)
+├── screenshot.png         # screenshot mode
+└── scroll.mp4             # scroll-video / record-video mode
+```
+
+### `capture.json` shape
+
+```json
+{
+  "source": { "url": "https://example.com", "finalUrl": "https://example.com/" },
+  "mode": "scroll-video",
+  "viewport": { "width": 1280, "height": 720, "deviceScaleFactor": 2 },
+  "outputs": [
+    { "type": "video", "localPath": "<output-dir>/scroll.mp4", "fps": 30, "durationSec": 8.0, "frames": 240 }
+  ]
+}
+```
+
+For `screenshot` mode each entry is `{ "type": "image", "localPath": "...", "fullPage": false }`.
+
+## Notes on the deprecated `microsoft/playwright-cli`
+
+The original `microsoft/playwright-cli` repository was **archived (read-only) in January 2021** and is deprecated; its functionality was folded into the main `playwright` package. This skill therefore uses the maintained **Playwright Python** library directly. Pure screenshots are also available via `npx playwright screenshot`, but scrolling-video capture requires script-driven scrolling + frame assembly, which is what this skill provides.
+
+## Error handling
+
+- **Playwright not installed**: the script detects it and prints `pip install playwright && playwright install chromium`.
+- **Browser binary missing** (`Executable doesn't exist`): run `playwright install chromium` (or the chosen browser).
+- **ffmpeg not installed** (video modes): the script prompts `brew install ffmpeg`.
+- **URL rejected (exit 5)**: non-http(s) scheme, or the host resolves to a private/internal/metadata address. Use a public URL, or set `WEB_CAPTURE_ALLOW_PRIVATE_HOSTS=1` for a trusted intranet target.
+- **Navigation timeout**: the page took too long; try `--wait-until load` or a smaller `--wait-ms`.
+- **Blank / partial capture**: increase `--wait-ms`, or some sites block headless browsers.
+- **Network**: verify the URL is reachable from the runtime host.