reference-media-analysis

What it does

Analyzes reference media (images, video, audio, mood boards, etc.) to determine the production function rather than reproducing protected expression. It instructs the agent to separate documented facts, production inferences, heuristics, and copying risks for safe adaptation. It requires starting with a clearance and risk gate, building a reference ledger with fields like reference_id, asset_type, source_location, permission_basis, license_evidence, allowed_uses, restricted_uses, and various risk flags. It provides triage levels Green, Yellow, Red/blocker and emphasizes provenance/metadata checks using C2PA, IPTC, EXIF/XMP, and file history. It distinguishes what to extract (concept, visual style, shot design, editing, performance, audio, product/brand, accessibility) from what not to copy. It instructs decomposing references by intent, audience, structure, craft, distinctiveness, transferable principles, risk, and production specs. It covers images, video, audio, avatar/face references with explicit guidance on what constitutes permissible analysis vs. copying. Finally, it details translating analysis into provider-independent prompts in a three-layer structure (production intent, abstracted learnings, safety/distinctness), with a prompt skeleton and a labeling scheme for reference assets, plus a similarity/plagiarism risk review scale and escalation triggers.

How it works

• For each reference, separate documented facts, production inferences, heuristics, and prohibited copying risks.
• Build a clearance and risk gate before deep analysis and create a minimum ledger with fields such as reference_id, asset_type, source_location, permission_basis, license_or_approval_evidence, allowed_uses, restricted_uses, people_present, likeness_or_voice_risk, provenance_checked, analysis_date, notes.
• Classify outputs with Green/Yellow/Red risk tiers and escalate if needed.
• Perform provenance checks using C2PA, IPTC, EXIF/XMP, and file history; record unknown/conflicting provenance as needed.
• Decompose references into universal categories (intent, structure, craft, distinctiveness, transferables, risk, production spec) and extract specific details for images, video, audio, and avatars.
• Create objective descriptions before interpretation, yielding a source description and production interpretation stored separately.
• Translate analysis into a three-layer prompt handoff:
  1. Production intent
  2. Abstracted reference learnings
  3. Safety and distinctness constraints
• Include a prompt skeleton with sections for Intent, Reference-derived direction, Originality constraints, Approved exact elements, Creative spec, Negative constraints, Disclosure/provenance.
• When accepting references, label each reference by function (e.g., ref_style_palette, ref_product_fidelity, ref_pacing, ref_performance_energy, ref_brand_asset).
• Provide a similarity/plagiarism risk review ranging 0–4 with escalation cues.

When to use it

Used when an agent must analyze reference media and translate insights into safe, non-copying direction, prompts, QA criteria, provenance records, and handoff notes for subsequent media production agents. It is specifically activated when production decisions hinge on understanding reference assets and ensuring rights, likeness, and platform disclosures are properly managed.

What it can touch

• Tools declared in the skill: claude-code, codex, copilot, cursor. (Provenance checks and ledger fields reference these workflows and tooling where applicable.)

Caveats

• License: MIT
• It emphasizes that rights, likeness, voice, trademark, privacy, or jurisdictional concerns may pause production and require client/legal approval. It discourages copying distinctive expression and mandates translation into new work preserving only approved product/brand details.
• It requires clear separation of observable facts from inferred production intent and cautions about provenance/metadata limitations and escalation when authenticity matters.
• No promises of outcomes; guidance centers on process and safety constraints as stated in the skill.