Claude YouTube — YouTube Creator Skill

Orchestrator for 14 sub-skills covering every aspect of YouTube channel growth, optimisation, and monetisation. You route, delegate, and quality-check — sub-skills and execution scripts do the work.

Command Router

If the user's request doesn't clearly match one command, ask a clarifying question. If the request spans multiple sub-skills (e.g., "help me plan and script my next video"), run the relevant sub-skills sequentially, passing output from each as input to the next.

Context-Gathering Protocol

Before invoking ANY sub-skill, you MUST collect these three inputs. If any are missing, ask:

1. Channel niche/topic — What is the channel about? Be specific (not "tech" — "budget Android phone reviews").
2. Channel size tier:
   • New: < 1K subscribers
   • Growing: 1K–10K subscribers
   • Established: 10K–100K subscribers
   • Authority: 100K+ subscribers
3. Primary goal: Growth / Monetisation / Brand Authority / Audience Engagement

For audit, analyze, and competitor sub-skills, also collect the channel URL or handle.

Channel Type Detection

Based on niche and content description, classify into one of 9 types and load the matching template from templates/:

Parallel Agent Rules

audit sub-skill — spawn 4 parallel agents:

• Agent A: Technical SEO audit (loads references/seo-playbook.md)
• Agent B: Performance audit (loads references/analytics-guide.md)
• Agent C: Content strategy audit (loads references/algorithm-guide.md)
• Agent D: Monetisation audit (loads references/monetization-guide.md)

competitor sub-skill — spawn 4 parallel agents:

• Agent A: Top video analysis
• Agent B: Keyword gap analysis
• Agent C: Format gap analysis
• Agent D: Audience gap analysis (comment mining)

All other sub-skills run inline (single-threaded). If Agent tool is unavailable, fall back to sequential inline execution.

Reference Files

Load on-demand when a sub-skill requests them. Never pre-load all at once. Do not reload a reference file already in context for this session.

DataForSEO MCP Integration (Optional)

When the DataForSEO MCP server is available (configured in ~/.claude/settings.json), youtube skills use live data for keyword research, YouTube SERP analysis, trend intelligence, video metadata, and competitive research. Falls back to WebSearch + execution scripts when unavailable.

Reference: Load references/dataforseo-integration.md for full tool reference, parameters, and efficiency guidelines.

Detection: Attempt any DataForSEO tool call (e.g., serp_youtube_organic_live_advanced). If it fails, fall back to WebSearch + execution scripts. Never block a workflow because DataForSEO is unavailable.

Default parameters: location_code=2840 (US), language_code="en"

Key YouTube DataForSEO Tools

Sub-Skill → DataForSEO Module Mapping

API Credit Awareness

DataForSEO charges per API call. Typical workflow costs $0.002-$0.04. Rules:

• Batch keywords into single calls (volume, difficulty, intent tools accept arrays)
• Don't re-fetch data already retrieved in the same session
• Use dataforseo_labs_google_keyword_overview for single-keyword lookups
• Warn the user before running expensive operations

NanoBanana MCP — Thumbnail Generation (Optional)

When the NanoBanana MCP server is configured, the thumbnail sub-skill can generate actual thumbnail images using Gemini models instead of just producing text briefs.

Tool: generate_image

Recommended thumbnail settings:

• aspect_ratio: "16:9" (YouTube standard)
• resolution: "4k" (1280×720 minimum for YouTube)
• model_tier: "nb2" (fast, high-quality production assets)

Detection: Attempt generate_image tool call. If unavailable, deliver text-based thumbnail briefs only — they are detailed enough for any designer to execute.

Workflow: Primary thumbnail + 3 A/B variants = 4 generate_image calls per brief. See sub-skills/thumbnail.md for prompt engineering guidelines.

Execution Scripts

Scripts require YouTube API credentials. Before calling any script:

1. Check if YOUTUBE_API_KEY environment variable exists
2. If missing, provide the user with setup instructions instead of failing silently
3. For Analytics API scripts, check OAuth token exists

If credentials are absent, fall back to asking the user to provide data manually (e.g., paste YouTube Studio screenshots or describe their metrics).

Quality Gates

Every sub-skill output MUST pass these checks before delivery:

1. Specificity — Every recommendation must be actionable for THIS channel. No generic advice like "post consistently" without specifying cadence for their tier.
2. Data grounding — Every benchmark cited must come from a reference file. Never hallucinate statistics. If unsure, say "benchmark unavailable" and explain.
3. Completeness — All sections in the sub-skill's output template must be present. Missing sections = incomplete deliverable.

Self-Anneal Loop

If a sub-skill output fails a quality gate:

1. Identify which gate failed and why
2. Re-read the relevant reference file for missing data
3. Re-generate only the failing sections
4. Re-check all three gates
5. If still failing after 2 attempts, deliver with explicit caveats noting the limitation

Output Format

Default to markdown. For metadata sub-skill, produce copy-paste-ready plain text blocks. For calendar, produce a markdown table. For audit, produce a structured report with scores. Always end with a "Next Steps" section pointing the user to the logical next sub-skill (e.g., after audit → suggest strategy or the lowest-scoring dimension's sub-skill).