Skip to main content
skill CX Codex

reference-decode — Claude Code skill: what it does, how to install

reference-decode turns a reference clip, frame, or contact sheet into prompt structure: hook essence, viewer question, visual grammar to copy, and imitation boundaries.

no license 1,000 installs ★ 26 updated 2026-10-31

Install into Claude Code · Skills CLI (Vercel) All methods and platforms ↓
npx skills add PostPlusAI/postplus-skills --skill reference-decode -a claude-code
Open on GitHub

What reference-decode does

The skill reads a benchmark video or frame set and extracts four objects: hookEssence (the opening mechanism), viewerQuestion (the question that holds attention), mustCopyVisualGrammar (what to repeat: shot order, object logic, timing), and forbiddenDrift (what not to copy — faces, wardrobe, exact location). Without a reference, it runs in proxy mode: builds structure from a segment pattern and the brief.

The process starts with the smallest set: the first 0–5 seconds of a clip, a hook contact sheet, or a note. The skill separates grammar from identity: it keeps camera work, order, object relationships; it drops exact appearance details. The output is a compact artifact passed to image-batch-runner or video-batch-runner before generation. Paid operations need PostPlus credits; without balance, the skill stops.

Works in Claude Code, Codex, Cursor, GitHub Copilot, Gemini CLI, and OpenCode. Install with npx skills add PostPlusAI/postplus-skills --skill reference-decode -a claude-code, and the file lands in ~/.claude/skills/reference-decode/SKILL.md.

What it helps with

  • break down a reference clip into hook, viewer question, and visual grammar
  • separate frame structure from specific faces and details before prompting
  • build prompt structure from a brief when no reference exists
  • pass grammar to image-batch-runner or video-batch-runner without extra artifacts

Example prompts for the agent

  • Take the first five seconds of ref/hook.mp4 and pull out the hook, viewer question, what to copy, and what not to.
  • From the contact sheet in board/intro-frames.png, assemble prompt structure: opening essence, frame grammar, and boundaries.
  • No brief, but I need a prompt for a three-answer segment: build a proxy structure with scene anchors.

How to install reference-decode

1Skills CLI (Vercel)

npx skills add PostPlusAI/postplus-skills --skill reference-decode -a claude-code

2Manually: put the files in place

~/.claude/skills/reference-decode/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/reference-decode/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about reference-decode

How to install reference-decode in Claude Code?

Run npx skills add PostPlusAI/postplus-skills --skill reference-decode -a claude-code. The skill will appear in ~/.claude/skills/reference-decode/SKILL.md.

What does reference-decode do?

It reads a reference clip, frame, or contact sheet and extracts hook essence, viewer question, visual grammar (what to repeat), and imitation boundaries (what not to copy). The result is passed to generation as prompt structure.

What does forbiddenDrift mean in the skill output?

A list of what not to copy from the reference: faces, exact wardrobe, location, overlays, creator identity. The skill separates structure (shot order, timing, object logic) from details that would make the result plagiarism.

Similar skills

All skills for AI and ML →

From the same set25

The whole set →
tiktok-research The tiktok-research skill gathers public TikTok material, from videos and comments to creator profiles and ad examples. Use it to examine audience language or compare content ideas. CX Codex +1 1,100installs youtube-research youtube-research — a skill that pulls public YouTube data: video metrics, channel summaries, and comment samples. Handy when you need evidence about content or audience. CX Codex +1 1,100installs facebook-research facebook-research gathers public Facebook posts, reels, comments, ad library listings, and events for audience or competitor research. CX Codex +1 1,100installs video-batch-runner video-batch-runner — a skill for Claude Code and Codex that runs video generation across six Seedance 2.5 modes via postplus CLI. It dispatches batch jobs from text, frames or audio and downloads finished clips. CX Codex +1 1,000installs image-batch-runner image-batch-runner — a skill for Claude Code and Codex that executes batch image briefs and edits. Runs jobs through the PostPlus CLI, monitors task status, and saves manifest records. CX Codex +1 1,000installs benchmark-to-brief benchmark-to-brief turns validated benchmark and audience evidence into campaign briefs, concepts, hooks, and test plans for the handoff from research to execution. CX Codex +1 1,000installs image-generation The image-generation Claude Code skill plans image creation and edits. It sorts the task, model, and reference rules before handing a ready request to the runner. CX Codex +1 709installs instagram-research instagram-research — a skill for gathering public Instagram data: profiles, posts, Reels, comments, hashtags, and creators. Use it for competitor audits, influencer discovery, and audience language analysis. CX Codex +1 579installs

Project sponsors

Companies that keep this analytics open