Poker Clip

Automatically cuts poker tournament videos into complete hand clips in vertical 9:16 format with subtitles and unique hooks for TikTok/YouTube Shorts.

Security Scan

The skill's stated purpose (cutting poker videos into short vertical clips) matches the included code, but the package metadata omits real dependencies, several SKILL.md instructions reference files that are missing, and there are sloppy/buggy bits in the code — these inconsistencies warrant caution before installing or running.

Purpose & Capability

The skill claims to require no binaries or credentials, but README and the scripts clearly expect Python 3.10+, ffmpeg on PATH, and the 'whisper' transcription library. That mismatch between declared requirements (none) and actual needs is a coherence problem.

Instruction Scope

Runtime instructions are narrowly scoped to local video transcription, segmentation, subtitle and hook overlay, and editing code to add signals. However SKILL.md and README reference additional helper scripts (analyze_hands.py, check_overlap.py, debug_signals.py, fix_hooks.py, check_clip5_boundary.py) that are not present in the file manifest; the instructions also suggest using a larger transcription model ('large-v3') without clarifying whether that is local or cloud-based. No external network exfiltration endpoints are present in the code.

Install Mechanism

There is no install spec (instruction-only), so nothing is downloaded automatically. The included scripts will write files into the workspace and call ffmpeg and whisper locally. This is a relatively low install risk, but the manifest omission of ffmpeg/whisper is misleading.

Credentials

The registry lists no required environment variables or credentials and the code does not read secrets, which is appropriate. But the README and SKILL.md imply large transcription models may be used (local heavy models or possibly cloud-hosted variants), and that would require significant local resources or API keys — the skill does not declare or document that tradeoff. The mismatch between declared zero env/deps and actual runtime needs is the core proportionality concern.

Persistence & Privilege

The skill does not request always:true and does not modify other skills or system-wide configuration. It does include a fix_paths.py script which edits files within the skill to convert hard-coded absolute paths into dynamic ones — that modifies local skill files but not outside resources. This is limited scope but worth noting.

What to consider before installing

This skill appears to do what it says (detect hands, transcribe, cut clips, overlay subtitles/hooks) but contains inconsistencies and some buggy/sloppy code. Before running it: 1) Don’t run it on sensitive videos or on a machine with sensitive data. 2) Confirm and install required software: Python 3.10+, ffmpeg on PATH, and the whisper transcription package (pip install openai-whisper or equivalent). 3) Inspect the scripts locally — several helper scripts the docs mention are missing, and poker_clipper.py contains at least one coding mistake (an undefined variable in the trailing-hand code path), so expect bugs. 4) If you need larger transcription models, ask the maintainer whether those are local models or cloud APIs (which would require API keys). 5) Run the skill in an isolated/test workspace (or VM/container) first. If you’re unsure, request the maintainer to: update registry metadata to list ffmpeg/whisper as requirements, provide the missing helper scripts (or remove references), and fix the obvious code issues before you install.

Poker Clipper Skill

用途

将长篇扑克比赛视频自动切割成完整手牌片段，转换为9:16竖屏格式，添加专业字幕和Hook文案，输出适合TikTok/YouTube Shorts的短视频。

调用方式

/poker-clipper [视频文件路径]

或直接说："帮我处理这个扑克视频" / "切割新视频"

核心脚本

脚本	作用
scripts/poker_clipper.py	主流程：转录→手牌检测→切割+字幕
scripts/gen_hooks.py	生成唯一Hook文案
scripts/patch_hooks.py	将Hook叠加到视频顶部黑区
scripts/fix_hooks.py	手动覆盖Hook（出现重复时用）
scripts/analyze_hands.py	诊断：显示完整街道/结果时间线
scripts/check_overlap.py	诊断：显示clip时间范围和间隔
scripts/debug_signals.py	诊断：显示转录中出现的结束信号

标准工作流

# 1. 放视频到 downloads/
# 2. 运行主流程
python scripts/poker_clipper.py "downloads/视频名.mp4"
# 3. 生成Hook
python scripts/gen_hooks.py
# 4. 叠加Hook
python scripts/patch_hooks.py
# 5. 打开查看
explorer clips

手牌边界检测原理

德州扑克手牌结构（必须理解）

Pre-flop → Flop（3张公牌）→ Turn（第4张）→ River（第5张）→ 摊牌/结果

每手牌是一个完整序列。切割必须从手牌开始到底池归属确认。

检测逻辑

1. 扫描转录，识别街道信号（Flop/Turn/River）和结果信号
2. 结果信号触发手牌结束，下一手从这里之后开始
3. 20s窗口去重：同一手牌结束时多个信号合并为一个
4. 评分：按兴奋关键词密度打分，取Top N手
5. 短片段过滤：<30s的片段自动丢弃

关键教训：哪些词不能做结束信号

• ❌ flop、turn、river — 这是街道词，出现在手牌中间
• ❌ full house、four of a kind — 解说在手牌进行中就会说
• ❌ all in、shoves — 行动词，不是结果
• ✅ 只有底池归属和玩家离开才是真正结束

结束信号库（HAND_END_SIGNALS）

持续维护，新信号随校准追加：

底池归属

• wins the pot, wins the hand, takes the pot, takes it down
• scoops, ship it, well done, nice hand, good hand
• win a pot worth, going to win a pot, raked in, rakes in
• locked up, two thirds of the way to scooping
• wins a monster, wins a massive, round one goes to

玩家弃牌/离场

• eliminated, busted, knocked out
• ronnie's folded, going to the cage
• folded his hand, folds his hand
• and he folds, and she folds

赛后评论（手牌已结束）

• phil deserved that, show some class, you lose the whole lot
• shuts me up, shut me up, locked up at least

画布参数（固定，勿改）

画布: 1080x1920（9:16竖屏）
视频区域: 1080x607，垂直居中（letterbox，完整保留横屏画面）
字幕区: SUB_Y=1383（底部黑区），44px字体
Hook区: HOOK_Y=328（顶部黑区），52px字体，前3秒显示

校准流程

当用户说"X分X秒开始是新的一手牌"：

1. 计算绝对时间：clip开始时间 + 用户报告的分秒
2. 运行 check_clip5_boundary.py（或临时脚本）查看该时间点前后转录
3. 找到解说中表示"上一手结束"的词
4. 将该词加入 HAND_END_SIGNALS
5. 重跑完整流程验证

Hook生成原则

• 5个公式轮换（每clip用不同公式）：悬念gap / 反直觉 / 情绪触发 / 部分揭示 / 沉浸感
• used_hooks set防止重复
• 金额从转录中提取（>100的数字才算）
• Fallback池保证不重复

常见问题

Q: 某手牌被切成两段 → 运行 analyze_hands.py 查看时间线，找中间误触发的结束信号，删掉它

Q: 两手牌被合并成一段 → 找两手牌之间的转录词，加入 HAND_END_SIGNALS

Q: Hook重复 → 运行 fix_hooks.py 手动赋值，然后运行 patch_hooks.py

Q: 字幕位置不对 → 调整 poker_clipper.py 中的 SUB_Y（字幕）或 HOOK_Y（Hook）

Q: Whisper识别错误导致信号丢失 → 在诊断脚本中扩大搜索范围，找到实际词汇加入信号库；或升级到 large-v3 模型（慢但准）

工作目录

C:\Users\user\.openclaw\workspace-poker\
├── downloads/     # 放输入视频
├── clips/         # 输出clip + 转录缓存 + JSON报告
└── scripts/       # 所有处理脚本