
Handoff Check|交件版本與審核檢查
交件前,確認檔案版本對得上
Check that delivery files match approvals.
US$5 一次買斷 / One-time download · Python 3.10+ · 可讀原碼 / Readable source
給剪輯師、工作室與交件管理者:找出漏檔、錯版,以及沒有對上版本的審核回條。
For editors, studios and delivery managers: catch missing files, wrong versions and mismatched review receipts.
先看成果 / See the result
合成範例 / Synthetic text fixtures
| 輸入 Input | 結果 Result | 意思 Meaning |
|---|---|---|
| 檔案與接受清單一致 / Matching files and manifest | PASS_INTEGRITY_ONLY | 檔案對得上 / Files match declarations |
| 故意填錯檔案指紋 / Deliberately wrong hash | BLOCKED | 阻擋錯版交件 / Mismatch blocks handoff |
三步開始 / Three steps
- 下載解壓,準備檔案與已接受的清單。 / Unzip and prepare files plus the accepted manifest.
- 執行下方合成示範,再檢查自己的交件。 / Run the demo below, then check your delivery.
- 依報告補檔或確認版本;PASS 不代表影音品質通過。 / Resolve reported issues; PASS is not a media-quality verdict.
English — Check your handoff before sending it
Handoff Check helps editors, studios and agencies find missing files, changed cuts and review receipts tied to the wrong version. It checks local file SHA-256 hashes, required delivery files and existing technical, visual and source-review receipts. You receive a JSON or Markdown report naming blockers and version differences.
US$5 one-time Download purchase · v1.1.1 Beta · English + Traditional Chinese instructions.
1. Before you start
You need a computer with Python 3.10+ and permission to read the delivery folder. macOS / Python 3.12.14 is tested; Python 3.10, Windows and Linux execution remain unverified. No FFmpeg, paid API key or extra Python package is needed for the standalone checker. An AI assistant is optional; its own fees and data-sharing rules apply.
The tool expects an accepted JSON manifest: a list of the approved files, their expected SHA-256 hashes and existing review receipts. It does not automatically review an MP4 just because you drop it into a folder.
2. Download and open the tool folder
Download the ZIP from your purchase and extract it into a new folder. Open a terminal in the extracted production-handoff-check folder. You should see SKILL.md, scripts and references. On macOS, type cd followed by a space, drag the folder into Terminal, then press Enter. On Windows, open PowerShell in that folder; if Python Launcher is installed, replace python3 below with py -3.
Check your Python version:
python3 --version
3. Run the included synthetic demo
Run these commands one at a time:
python3 scripts/demo.py demo-package
python3 scripts/audit.py demo-package/manifest.json --root demo-package --format markdown
python3 scripts/audit.py demo-package/blocked-manifest.json --root demo-package
The first command creates synthetic files in a NEW demo-package directory. If that directory already exists, use a fresh extraction or another new directory; do not overwrite an earlier job.
The second command should show PASS_INTEGRITY_ONLY and exit 0. The third deliberately uses a wrong hash and should show BLOCKED and exit 2. That blocked result means the demo worked. These are synthetic text fixtures, not real video or genuine QA evidence.
4. Check your own delivery
- Ask the responsible producer for the accepted artifact, work-order ID, revision, expected hashes and existing review receipts.
- Follow references/input.md to prepare your manifest.json. The technical, visual and source review kinds are all required; review statuses must be the exact string PASS. Receipt fields can be mapped with the documented JSON Pointers.
- For contracted cover, subtitle or caption files, add required_roles and deliverables as described in references/quickstart.md.
- Keep files in a stable local delivery folder. Manifest file paths are relative to that folder; traversal and symlinks are rejected.
- Run the command below, replacing both quoted placeholders with your actual paths:
python3 scripts/audit.py "path/to/manifest.json" --root "path/to/delivery-folder" --format markdown
Omit --format markdown to receive JSON. If you want to save the report, append > "handoff-report.md"; this writes a report in your current folder and replaces that file if it already exists. Choose a new report name for each revision.
Optional AI request: “Use the local production-handoff-check SKILL.md to check this accepted manifest and delivery folder. Explain each blocker. Do not change the files, replace expected hashes, invent reviews or publish anything.”
5. Read the result and fix the cause
- PASS_INTEGRITY_ONLY: the files match the supplied declarations. The responsible person decides the next step.
- Missing file: check the reported relative path and package root, or restore the accepted file.
- SHA mismatch: the file bytes differ. Locate the accepted version or obtain a properly reviewed revision; never recalculate the expected hash merely to force a pass.
- Wrong revision / work order: use a review for the matching job and version.
- Missing or non-PASS review: obtain the required genuine review through your normal process. The tool cannot create one.
- Python not found: install Python from python.org or ask your technical contact. “Cannot open scripts/audit.py” usually means the terminal is in the wrong folder.
Scope, privacy and support
PASS does not prove reviewer authenticity, media quality, licensing or publication approval. Missing identity fields in a receipt cannot prove identity. No media generation/editing, OCR, audio QA or automatic publishing is included. Use stable local copies; this is not a hostile-filesystem sandbox.
The checker reads named local files and makes no network or paid API calls. The demo generator writes synthetic fixtures. If you share files with an AI client or support, that is a separate action subject to their policies.
Support: [email protected]. Includes installation guidance and reproducible bug reports; excludes custom integration, manual client-video review, rights clearance and 24/7 live service. Send the package version, OS, Python version, command and redacted error output. Use synthetic examples; do not send client media, identity documents or API keys. No fixed response time or perpetual update promise. Refunds follow current Capafy rules and applicable law.
Local evidence: 23 regression tests, 14 independent agent-run synthetic CLI checks and ZIP extraction replay passed on macOS; the official skill format validator passed. These are not real-user trials or Capafy certification. Capafy installation remains unverified.
繁體中文 — 交片前,先確認檔案與版本
Handoff Check 幫剪輯師、工作室與代理商找出漏檔、改過的成片,以及綁到舊版本的審核回條。工具在本機檢查 SHA-256、必交配件與既有技術/視覺/來源回條,提供 JSON 或 Markdown 報告,列出問題檔案及版本差異。
US$5 一次買斷下載 · v1.1.1 Beta · 英文與繁體中文指引。
1. 使用前準備
需要 Python 3.10+ 及交件資料夾讀取權限。目前實測 macOS/Python 3.12.14;Python 3.10 最低版、Windows/Linux 未實證。獨立命令列不需 FFmpeg、付費 API 金鑰或額外 Python 套件。AI 助理可選用,費用與分享政策另計。
需要一份已接受的 JSON manifest,列出核准檔案、預期 SHA-256 與既有審核回條。不是把 MP4 拖入資料夾就能全自動驗片。
2. 下載並進入資料夾
從購買項目下載 ZIP,解壓到新資料夾,在含 SKILL.md、scripts、references 的 production-handoff-check 資料夾開啟終端機。macOS 可輸入 cd 加空格,拖入資料夾後按 Enter;Windows 可在資料夾開啟 PowerShell,若使用 Python Launcher,將上方命令的 python3 改成 py -3。先執行 python3 --version 確認版本。
3. 先跑合成範例
依英文第 3 節的三行命令逐行執行:第一行建立新的 demo-package;第二行應得到 PASS_INTEGRITY_ONLY(退出碼 0);第三行故意填錯 SHA,應得到 BLOCKED(退出碼 2),表示成功攔截。
範例是合成文字檔,不是真實影片或 QA 證據。demo-package 已存在時,改用新解壓資料夾或新名稱,不覆蓋既有工單。
4. 檢查自己的交件
- 向負責人取得核准成片、工單 ID、版本、預期 SHA 與既有審核回條。
- 依 references/input.md 建立 manifest.json;technical、visual、source 三種回條都要有,狀態須為精確的 PASS。依規格指定回條欄位的 JSON Pointer。
- 必交封面、字幕、文案,依 references/quickstart.md 加入 required_roles 與 deliverables。
- 檔案放在穩定本機資料夾;manifest 的檔案路徑相對於此資料夾,不接受越界路徑或符號連結。
- 使用英文第 4 節命令,把引號內 manifest 路徑及 delivery-folder 路徑換成實際位置。省略 --format markdown 就輸出 JSON。
若要存報告,在命令後加 > "handoff-report.md"。這會寫入目前資料夾,若同名檔已存在會覆蓋,請每版使用新檔名。
可交給 AI 的指令:「使用本地 production-handoff-check 的 SKILL.md 檢查這份已接受 manifest 與交件資料夾,解釋阻擋原因。不要改檔、替換預期 SHA、捏造審核或發布。」
5. 結果與排除方式
- PASS_INTEGRITY_ONLY:檔案與給定聲明一致,由負責人決定下一步。
- 缺檔:檢查報告路徑、--root 資料夾,或補回已核准檔案。
- SHA 不符:檔案內容有差異,找回核准版本或取得新版本的正式審核;不要為了通過而重算預期 SHA。
- 版本/工單不符:換成同工單、同版本的回條。
- 缺少審核或狀態不是 PASS:循原流程取得真實審核,工具不能代產回條。
- 找不到 Python:由 python.org 安裝或請技術人員協助;找不到 scripts/audit.py 通常是終端機所在資料夾不對。
範圍、資料與客服
PASS 不保證回條真實、影音品質、素材授權或發布核准。回條缺少身分欄位時,不能證明工單身分。工具不製片、不剪輯、不做 OCR/音訊 QA、不自動發布,也不是對抗惡意檔案系統的沙箱。
檢查器只讀指定本機檔案,不呼叫網路或付費 API;demo 會建立合成檔案。另行分享給 AI 客戶端或客服時,依該服務規則處理。
客服:[email protected]。提供安裝指引與可重現的程式錯誤回報支援;不含客製整合、人工驗客戶影片、素材授權審核或 24 小時即時服務。請附商品版本、作業系統、Python 版本、命令及遮去私人路徑的錯誤文字,使用合成案例,不寄客戶影片、證件或 API 金鑰。不承諾固定回覆時間或永久更新;退款依 Capafy 當次規則及適用法令。
本機已有 23 項回歸測試、14 次獨立代理合成 CLI 與 ZIP 解壓重播通過,官方格式驗證亦通過;這不等於真人試用或平台認證,Capafy 安裝仍未實證。

