How to use the Video Shoot Logger agent skill (cheat-shoot)

What does the cheat-shoot skill make the agent do?

The skill moves a video from the state of having a written prediction but no filming to the state of being filmed but not published. It has a single job: record the shoot and keep the count of waiting videos accurate. It does not write the prediction content itself. Prediction files are written by the separate cheat-predict step.

When you log a shoot, the agent creates a video folder for it if one does not exist, asks whether the filmed script matches the planned script in scripts/, and adds the video to the shoots queue in the shared state file. Then it reports how many videos are waiting. If you changed the script a lot, the agent hands the video to the prediction step to score a second version, keeping the original prediction on file.

The skill is designed to pair with cheat-publish. Filming adds an item to the queue and publishing removes it.

• Checks that scripts/<id>.md exists and that a matching prediction file exists
• Refuses to log a shoot that has no prior prediction
• Creates videos/<id>/ and records the script consistency result
• Adds the video to state.shoots and reports the buffer count and colour

Install the cheat-shoot skill from the ZIP

Download the ZIP for the skill from this page and unzip it. You can install it in one of two places. For a personal install that works in every project, put the cheat-shoot folder in ~/.claude/skills/. For a single project, put it in that project's .claude/skills/ folder.

If you use the Claude apps, you can instead upload the skill under Settings → Capabilities → Skills.

The skill's folder contains SKILL.md. That file is the skill definition the agent reads. The listing describes the skill's files as SKILL.md only.

• Personal install: ~/.claude/skills/cheat-shoot
• Project install: .claude/skills/cheat-shoot inside your project
• Claude apps: Settings → Capabilities → Skills, then upload

Set up your cheat-on-content project

The skill works on a project folder that follows the cheat-on-content layout. The source expects a scripts/ folder with pre-shoot drafts, a predictions/ folder with prediction files, and a .cheat-state.json state file in the project. Each script and prediction pair shares the same name, and the skill finds them by id.

The skill creates a videos/ folder for each filmed video as it runs. You do not need to create that folder yourself.

• scripts/ holds pre-shoot drafts named with the video id
• predictions/ holds the v1 prediction for each video, written before filming
• videos/ receives one folder per filmed video, created by the skill
• The state file .cheat-state.json tracks the shoots list

Trigger the skill with natural requests

You do not need to type a command. The skill's description lists the phrases that trigger it. In Chinese these include 「拍了」, 「拍了 X」, 「已拍 X」 and 「录完了」. In English the listed triggers are shot and shot it. The skill also accepts the argument form /cheat-shoot followed by a script path or id.

You can give a full path, a shortened name such as the date and title part of the file name, or just the id. The agent uses a glob search to find the matching script when you give only the id. If you log several videos in a row, say each one in turn, for example 「拍了 X」 then 「拍了 Y」 then 「拍了 Z」.

• Full path: scripts/<id>_<title>.md
• Short name: the file name without the folder and extension
• Id only: the agent searches scripts/ for a match
• Batch: several shoots in one session, logged one at a time

Answer the script consistency question

After the folder is created, the agent asks whether the script you filmed matches the planned draft. There are three answers. Choose a if you filmed the draft as written. Choose b if you changed some of it, and paste the filmed script, give a file path, or give a transcript. Choose c if the video is essentially a different script, in which case the agent points you to a redo flow that writes a new script and a new prediction.

For option b, the agent measures how much the text changed. The skill's threshold is 30 percent after normalising the text. At or above that level, it proposes a v2 prediction. Below it, it asks whether you still want one. If you improvised and kept no script, the agent marks the shoot as script lost and warns you to keep a script or transcript next time.

• a: consistent, no re-scoring
• b: modified, diff measured, v2 offered at 30 percent or more
• c: redo flow, the skill stops without adding to the queue
• No script kept: marked as lost, v2 skipped

Read the state and buffer output

After logging, the agent reports the new video folder and its prediction file. It then reports the buffer count and a colour. The colour depends on your publishing cadence rules in the shared cadence protocol. If the buffer moves to a higher level, the agent warns that the backlog is growing and suggests pausing filming to publish and review.

Each entry in state.shoots records the video folder, prediction file, scripts path, shot time, script consistency, diff percentage, whether a v2 prediction was written, and a short hash of the script at shoot time. The list is kept in shot-time order so the oldest unpublished video is first.

• script_consistency: consistent, modified or lost
• script_diff_pct: a whole number from 0 to 100, or null
• v2_prediction_written: true when the prediction file has a v2 section
• shot_at: ISO timestamp, oldest first

Safety notes: scripts and network access

The listing says the skill runs scripts and does not use the network. The skill's allowed tools include shell commands, file reads and writes, edits and file search. It uses Python 3 when available to run tools/diff_pct.py from the cheat-on-content repository, which it locates by resolving a symlink from ~/.claude/skills/cheat-shoot back to the repository. If python3 or that tool is not found, it falls back to a line-based diff using the standard diff command.

This means the skill's diff measurement depends on the install method. A symlink to the repository gives the normalised character-level diff. A plain unzipped copy with no link to the repository will use the less accurate line-level fallback, and the agent warns you when this happens. Review the script and the allowed tools before you install it, as with any skill that can run commands.

• Runs scripts: yes, through shell commands the agent executes
• Network access: none stated in the listing
• Optional Python 3 and rapidfuzz; falls back to difflib or the shell diff
• Diff tool is found through a symlink to the cheat-on-content repository