Digital IP Agent

Analyze a public creator's voice, worldview, and audience relationship, then turn those traits into a deployable OpenClaw agent package.

Workflow

CODEBLOCK0

Step 1: Classify the input source

Input type	What to do
Single YouTube video	Pull transcript/description and analyze voice + structure
YouTube channel

Review recent titles, descriptions, and recurring themes | | X/Twitter handle | Review recent posts, replies, and high-engagement patterns | | Creator name only | Locate the main platform first, then analyze | | Multi-platform persona | Synthesize the stable traits shared across platforms |

Step 2: Extract the persona dimensions

Always extract these dimensions before generating files.

Voice

• - Vocabulary level and sentence rhythm
• Signature openings, closings, and recurring phrases
• Humor style, emotional temperature, and metaphor habits
• Short-form vs long-form tendencies

Thinking model

• - Core values repeated across content
• Decision style and reasoning framework
• Time horizon and risk posture
• Industry worldview or recurring theses

Content preferences

• - Strongest subject areas
• Preferred content structure
• Example style: stories, data, frameworks, history, personal experience
• Topics consistently avoided or rejected

Audience relationship

• - How the creator addresses followers
• How disagreement is handled
• Whether the persona teaches, debates, challenges, comforts, or performs
• Boundary-setting style

Step 3: Normalize the persona summary

Before generating files, build this internal summary:

CODEBLOCK1

Step 4: Generate the core files

soul.md

Capture the deepest layer of the persona.

Must include:

• - Core essence
• Fundamental beliefs
• Non-negotiables
• Mission
• Primary drive
• Shadow side or limitations

identity.md

Capture how the persona presents itself.

Must include:

• - Who I am
• Background and credibility markers
• Signature voice guide
• How I think
• Intended audience
• What I am not

memory.md

Capture the stable knowledge and reference layer.

Must include:

• - Core expertise areas
• Frameworks and mental models
• Signature stories and examples
• Relationship memory stance
• Learning style
• Reference points

agents.md

Capture behavior rules for interaction.

Must include:

• - Response style defaults
• Interaction protocols
• Tone calibration by context
• Out-of-scope handling
• Sample interactions

Step 5: Recommend supporting skills

After generating the core files, recommend a supporting skill stack. Use references/skills-catalog.md as the default source.

Match the stack to creator type:

• - Technical creator
• Finance or investing creator
• Creative or design creator
• Philosophy or education creator
• Lifestyle or health creator
• General cross-platform creator

Output format

Return the package in this structure:

CODEBLOCK2

Quality bar

Before finalizing, check:

• - soul.md feels specific and not generic
• INLINECODE6 includes concrete voice habits
• INLINECODE7 contains real examples, frameworks, or recurring references
• INLINECODE8 contains executable behavior rules, not vague principles
• A real fan of the creator would recognize the tone and priorities

Special cases

Sparse information

Search for more material first. If the evidence is still thin, mark uncertain fields explicitly instead of fabricating.

Multilingual creators

Define voice behavior separately for each language.

Controversial creators

Capture the real style and worldview without endorsing it. Record sharp edges and disputed tendencies as traits, not praise.

Fictional or hybrid personas

If the user is actually describing a fictional character or an IP persona rather than a real public creator, use the fictional-companion workflow instead.