Content Setup

Content Setup Skill

When This Skill Activates

This skill runs in three scenarios:

1. Explicit invocation: user runs /content setup, /content brand show, /content brand update, or /content voice.
2. Missing brand detection: any other content skill (e.g., /content draft, /content social) checks for the brand/ directory and it does not exist. That skill should hand off here before proceeding.
3. Voice recalibration: user drops writing samples into brand/voice-samples/ and runs /content voice to re-derive voice attributes.

Command Routing

The Onboarding Philosophy

This is a conversation, not a form. Claude should:

• Ask one question at a time (sometimes two if they are tightly related).
• Listen to the answer, then probe deeper when the response is vague or generic.
• Challenge cliches. If the user says "we empower people," ask what that looks like on a Tuesday afternoon.
• Reflect back what you heard before moving to the next phase, so the user can correct misunderstandings early.
• Keep a running internal summary; do not ask the user to repeat themselves.
• Be warm but direct. This process should feel like working with a sharp brand strategist, not filling out a government form.

Never dump all questions at once. The flow is adaptive -- skip questions whose answers are already obvious from prior responses. Add follow-ups when something interesting surfaces.

Phase 1: Identity (5-7 questions)

Goal: Understand who this brand is at its core.

Key topics:

• Brand or company name
• What you actually do (product/service/mission)
• Who you serve (audience -- push for specifics)
• The elevator pitch (30 seconds or less)
• Competitors and how you differ from them
• The single biggest differentiator
• One-word brand essence

Probe hard on differentiators. "Better quality" is not a differentiator. "We hand-test every unit before shipping" is.

See questions/brand-identity.md for the full question script with follow-up probes.

See questions/audience.md for the detailed audience deep-dive.

Phase 2: Voice (4-5 questions)

Goal: Capture how the brand sounds when it writes.

Key topics:

• Voices they admire (brands, writers, publications)
• Formality scale from 1 (SMS to a friend) to 10 (Supreme Court brief)
• Humor style (dry, playful, none, self-deprecating, absurdist)
• Favorite phrases and words they love to use
• Forbidden words and phrases (what should never appear)
• Writing samples (if available, analyze them)

If writing samples exist in brand/voice-samples/, analyze them for patterns: sentence length, vocabulary level, punctuation habits, rhetorical devices, tone shifts.

See questions/voice-tone.md for the full question script with the formality scale reference.

Phase 3: Visual (4-5 questions)

Goal: Document the visual identity so content can reference it.

Key topics:

• Logo description and usage rules
• Primary and secondary color palette (hex values if known)
• Reference brands whose visual style they admire
• Typography preferences (serif vs sans-serif, specific fonts)
• Photography and illustration style (candid vs posed, illustrations vs photos, abstract vs literal)

Claude is not designing the brand -- just documenting what exists or what the user aspires to. If they have no visual identity yet, capture their preferences and inspirations.

See questions/visual-style.md for the full question script.

Phase 4: Platforms (3-4 questions)

Goal: Know where the content lives and how often it ships.

Key topics:

• Current publishing platforms (blog, social, newsletter, podcast, etc.)
• Posting frequency goals per platform
• Content types per platform (long-form, short-form, threads, video scripts)
• Platforms they explicitly want to avoid and why

See questions/platforms.md for the full question script.

Phase 5: Review

Goal: Confirm everything before generating files.

Process:

1. Present a structured summary of all captured information, organized by phase.
2. Ask: "What did I get wrong? What's missing?"
3. Iterate until the user confirms.
4. Generate the output files.

Output Generation

After the user confirms the review, generate three files using the templates:

Create the brand/ directory if it does not exist. If files already exist, confirm before overwriting.

See the templates/ directory for the exact output format of each file.

/content brand show

When the user runs /content brand show:

1. Check that brand/BRAND.md, brand/VOICE.md, and brand/VISUAL.md exist.
2. If any are missing, report which files are missing and offer to run setup.
3. If all exist, read each file and present a clean summary:
   • Brand identity in 2-3 sentences
   • Voice profile (key attributes, formality level)
   • Visual snapshot (primary colors, typography)
   • Active platforms and frequency
4. End with: "Run /content brand update to modify any of this."

/content brand update

When the user runs /content brand update:

1. Read all existing brand files.
2. Ask: "What would you like to change?" Accept free-form input.
3. Map the requested changes to the relevant file(s).
4. Show a diff-style preview of what will change.
5. Confirm before writing.
6. Update only the affected sections, preserving everything else.

This is not a full re-onboarding. It is a surgical edit. If the user wants to start over, they should run /content setup again.

/content voice

When the user runs /content voice:

1. Check for brand/voice-samples/ directory.
2. If it does not exist, create it and ask the user to drop 3-5 writing samples there (blog posts, emails, social posts -- anything representative).
3. If samples exist, analyze them for:
   • Average sentence length and variation
   • Vocabulary complexity (Flesch-Kincaid or similar heuristic)
   • Punctuation patterns (em-dash heavy? Semicolons? Exclamation points?)
   • Rhetorical devices (questions, lists, metaphors, analogies)
   • Tone markers (formal/casual, confident/tentative, warm/distant)
   • Signature phrases or recurring patterns
4. Present findings to the user for confirmation.
5. Update brand/VOICE.md with the analysis, merging with any existing manual entries.

Error Handling

• If the user abandons mid-flow, save progress to brand/.setup-draft.json so they can resume later.
• If brand files exist when /content setup is invoked, warn the user and ask whether to start fresh or update.
• If writing samples are too short (under 200 words total), warn that the voice analysis will be unreliable and ask for more material.

Quality Bar

A completed brand setup must pass the validation checklist in validation.md. Key requirements:

• No placeholder text or "TBD" entries in final output
• Audience description includes at least: role, experience level, and primary pain point
• Voice section includes at least 3 "sounds like us" and 3 "doesn't sound like us" examples
• At least one publishing platform defined with a frequency goal
• Differentiator is specific and falsifiable (a competitor could not claim the same thing)