model-release-briefing

name: model-release-briefing description: Use when a new Anthropic model is released and the user wants a personalized briefing and hands-on tutorial. Use when the user says "new model released", "walk me through the release", "briefing on Claude X", or invokes /model-release-briefing. user-invocable: true argument-hint:

Model Release Briefing

Generate a personalized briefing and hands-on tutorial for a new Anthropic model release. Six phases, executed sequentially. Each phase must complete before the next begins.

Phase 1: Ingest Source Materials

Read all files at the provided path ($ARGUMENTS[1], or ask the user). Accept blog posts, transcripts, summaries, prompt templates, changelogs, release notes. Build a comprehensive picture of every change in the release.

Phase 2: Gather User Context

Check conversation history and memory files for user context. If insufficient, ask these 4 intake questions and wait for answers before proceeding:

1. What is your role? (developer, PM, researcher, etc.)
2. How do you primarily use Claude? (coding, writing, analysis, agents, etc.)
3. Which Claude products do you use? (API, Claude.ai, Claude Code, etc.)
4. What is your most common AI workflow?

Phase 3: Generate Personalized Briefing

Using briefing-reference-template.md, walk through every category of changes. For each item: what it is, what it does in practice, why it matters to this specific user. Structure:

• Open with "What Matters Most for You" - top 3-4 changes ranked by impact for their role/workflow
• Walk through all categories - skip nothing, but weight detail by relevance
• Close with "Practical Takeaways" - 3-7 actionable items they can do today

Phase 4: Identify Tutorial Features

From the briefing, select the 3-5 features with highest practical impact for this user's workflows. Present the selection to the user and get confirmation before proceeding.

Phase 5: Create Self-Contained Tutorial

Using tutorial-structure.md, generate a hands-on tutorial. Requirements:

• Global prerequisites section (tools, subscriptions, CLI versions)
• Per-feature prerequisites (SDK versions, env vars, feature flags)
• Step-by-step exercises with working code
• Verification steps for each exercise
• Zero reliance on external docs - everything needed is inline

Phase 6: Verification Pass (Non-Negotiable)

For every feature flag, config value, API parameter, and setup step in the tutorial:

1. Fetch the official documentation (docs.anthropic.com, relevant SDK docs)
2. Verify each claim against the source
3. Fix any discrepancies before presenting to the user

Use verification-checklist.md as the verification guide. Do not skip this phase.

Lessons Learned

These are real errors caught in previous runs. Keep them in mind throughout:

• Features may be disabled by default. Example: agent teams required an env var (CLAUDE_CODE_ENABLE_TEAMS=1) not mentioned in blog posts. Always check for opt-in requirements.
• API parameters change between model versions. Example: budget_tokens for extended thinking was deprecated in favor of token_count with type: "adaptive". Always verify parameter names and types.
• SDK versions matter. New parameters often require the latest SDK version. Pin and document the minimum version.
• "Max" effort levels may be model-exclusive. Some settings (like thinking: {"type": "enabled", "budget_tokens": "max"}) only work on specific models.
• Subagent vs. agent team confusion is real. Always include a clear comparison table when both concepts exist.
• Blog posts and summaries contain errors. Never rely solely on marketing materials. Always verify against official API docs and changelogs.