feat: update speaker notes with capstone project examples

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>

Builders-Class-5-Slides.html

@@ -280,7 +280,7 @@
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">You've built RAG systems. That's context-aware answers — you ask, it reads your files, it answers. An agent goes further: it can use tools, take actions, make decisions, and chain steps without you directing each one. Today we're building agents — but starting from what you already know.</aside>
+  <aside class="notes">You've built RAG systems. That's context-aware answers — you ask, it reads your files, it answers. An agent goes further: it can use tools, take actions, make decisions, and chain steps without you directing each one. Today we're building agents — but starting from what you already know. Think about your capstones: a URL-to-video pipeline that turns articles into vertical video? That's an agent — it fetches, transforms, generates voiceover, adds subtitles, all autonomously. A political events calendar that scans sources and outputs structured data? Agent. Even the WhatsApp podcast recommender — it listens, reasons about preferences, and takes action. You're all building agents already, you just need the scaffolding.</aside>
 </section>
 
 
@@ -372,7 +372,7 @@
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">Skills cover 90% of journalism use cases — they're markdown files, no coding. MCP is the next level: connecting to live external data. Subagents are autonomous — they can run multi-step research on their own. Today: skills. Class 6: MCP live install.</aside>
+  <aside class="notes">Skills cover 90% of journalism use cases — they're markdown files, no coding. MCP is the next level: connecting to live external data. Subagents are autonomous — they can run multi-step research on their own. Today: skills. Class 6: MCP live install. Quick mental exercise: which block does your capstone need most? The headline refiner and journalist toolkit are pure skills — markdown instructions, no APIs needed. The video pipeline and cultural-references-to-streaming system need MCP to connect to external services. The Slack emoji game might use subagents to manage parallel game states. We'll get to all of these.</aside>
 </section>
 
 
@@ -492,7 +492,7 @@
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">This folder structure is what separates a generic chatbot from your personal assistant. AGENTS.md is identity. context/ is knowledge. skills/ is workflows. The AI can read all of it. The more context you give it, the more relevant its help becomes. Think of it as setting up a desk for a new colleague — but you control exactly what they can see.</aside>
+  <aside class="notes">This folder structure is what separates a generic chatbot from your personal assistant. AGENTS.md is identity. context/ is knowledge. skills/ is workflows. The AI can read all of it. The more context you give it, the more relevant its help becomes. Think of it as setting up a desk for a new colleague — but you control exactly what they can see. For your capstones, think about what goes in context/: if you're building the story archive with conversational interviews, your context/ folder might have your publication's archive structure and interview templates. If you're building the podcast recommender, it might have listener personas and genre taxonomies.</aside>
 </section>
 
 
@@ -529,7 +529,7 @@
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">Start filling this in now. The better you describe yourself, the more relevant every response becomes. Your beat, your role, your current projects, how you like things formatted. This is the difference between a generic AI and your AI. Take 5 minutes — I'll come around to help.</aside>
+  <aside class="notes">Start filling this in now. The better you describe yourself, the more relevant every response becomes. Your beat, your role, your current projects, how you like things formatted. This is the difference between a generic AI and your AI. Take 5 minutes — I'll come around to help. Concrete examples from your projects: if you're building the headline refiner, your AGENTS.md starts with your house style rules and what makes a good headline at your org. If you're building the cultural-references system, it starts with your broadcaster's streaming catalog structure. The "Current Projects" section should describe your capstone — that way the assistant can connect everything it does back to what you're building. MAP YOUR CAPSTONE: write the first 3 lines of your project's AGENTS.md right now.</aside>
 </section>
 
 
@@ -554,7 +554,7 @@
     <p style="opacity: 0.7; margin-top: 1em;">If it gets something wrong — that's your cue to improve the file.</p>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">This is the test. If it describes you accurately, your AGENTS.md is doing its job. If it misses something or gets it wrong, that's signal — go back and add the missing context. The iteration loop on AGENTS.md is as important as writing it the first time.</aside>
+  <aside class="notes">This is the test. If it describes you accurately, your AGENTS.md is doing its job. If it misses something or gets it wrong, that's signal — go back and add the missing context. The iteration loop on AGENTS.md is as important as writing it the first time. Bonus prompt for capstone teams: try asking "How would you help me build [your capstone project]?" — if the assistant gives generic advice, your AGENTS.md needs more project-specific context. If it gives targeted suggestions, you're on the right track.</aside>
 </section>
 
 
@@ -584,7 +584,7 @@
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">A skill is just a markdown file — a text file with instructions. Interview prep, fact-checking, daily digest, story pitching. You write it once in plain English: "When I invoke this skill, read my beat notes and draft 10 interview questions organized by theme." The AI follows the instructions every time. No code. Key insight: skills are shareable. Your interview prep skill works for everyone on your team.</aside>
+  <aside class="notes">A skill is just a markdown file — a text file with instructions. Interview prep, fact-checking, daily digest, story pitching. You write it once in plain English: "When I invoke this skill, read my beat notes and draft 10 interview questions organized by theme." The AI follows the instructions every time. No code. Key insight: skills are shareable. Your interview prep skill works for everyone on your team. MAP YOUR CAPSTONE: what's the first skill your project needs? If you're building the video pipeline, it's probably an "article-to-script" skill that takes a URL and outputs a video script with shot notes. The journalist toolkit? That's literally three skills in a trench coat — source-finder, interview-prep, timeline-builder. The headline refiner is one focused skill with your house style rules baked in.</aside>
 </section>
 
 
@@ -666,7 +666,7 @@ Save to drafts/daily-digest-YYYY-MM-DD.md:
     <p style="opacity: 0.7; margin-top: 1em;">Then test it: <em>"Give me my daily digest"</em></p>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">Everyone opens skills/daily-digest.md and starts editing. The goal isn't perfection — it's getting a version that reflects your actual workflow. If you cover courts, add "Check recent court filings." If you cover tech, add "Scan Product Hunt and Hacker News." After editing, test it: "Give me my daily digest." What worked? What's missing? Edit again. This is the loop.</aside>
+  <aside class="notes">Everyone opens skills/daily-digest.md and starts editing. The goal isn't perfection — it's getting a version that reflects your actual workflow. If you cover courts, add "Check recent court filings." If you cover tech, add "Scan Product Hunt and Hacker News." After editing, test it: "Give me my daily digest." What worked? What's missing? Edit again. This is the loop. Think about how a daily digest connects to your capstone: the political events calendar team could adapt this template to scan for political events instead of news. The cultural-references team could build a "what's trending in culture today" digest that feeds their recommendation engine. Same pattern, different content.</aside>
 </section>
 
 
@@ -839,7 +839,7 @@ When I ask you to prepare for an interview:
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">Walk through each step. The power is in step 1: it reads your actual beat notes, not generic journalism knowledge. Steps 6-7 are important — it connects research to your specific work and flags what to verify. After we build this together, the next slide is the test.</aside>
+  <aside class="notes">Walk through each step. The power is in step 1: it reads your actual beat notes, not generic journalism knowledge. Steps 6-7 are important — it connects research to your specific work and flags what to verify. After we build this together, the next slide is the test. Notice the pattern: trigger, steps, output format, rules. This is the same skeleton for every skill you'll build — including for your capstones. The story archive team could use this exact structure for a "story-intake" skill: when I want to archive a story, read my archive structure, ask me guided questions about the piece, output a structured record. Same four parts.</aside>
 </section>
 
 
@@ -863,7 +863,7 @@ When I ask you to prepare for an interview:
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">The iteration loop is the skill. Don't aim for perfect on the first run — aim for useful enough to improve. If it misses your org's focus, add that to AGENTS.md. If it misses a step, add it to the skill. Every refinement makes the next run better.</aside>
+  <aside class="notes">The iteration loop is the skill. Don't aim for perfect on the first run — aim for useful enough to improve. If it misses your org's focus, add that to AGENTS.md. If it misses a step, add it to the skill. Every refinement makes the next run better. This is exactly how you'll develop your capstone skills too. The Slack emoji game team won't get the game rules right on attempt one — but each failed round tells you what to add to the skill file. The headline refiner will suggest awkward headlines at first — add your house style gotchas and it gets sharper every iteration.</aside>
 </section>
 
 
@@ -897,7 +897,7 @@ When I ask you to prepare for an interview:
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">This isn't a tool you learn once and use forever unchanged. It's a system that compounds. The people who get the most from AI assistants are the ones who spend 10 minutes a week refining their setup — adding a beat note, sharpening a skill, updating their AGENTS.md with something new they learned. The best organizations share skills across teams.</aside>
+  <aside class="notes">This isn't a tool you learn once and use forever unchanged. It's a system that compounds. The people who get the most from AI assistants are the ones who spend 10 minutes a week refining their setup — adding a beat note, sharpening a skill, updating their AGENTS.md with something new they learned. The best organizations share skills across teams. For your capstones: every piece of research, every dataset, every source list you add to your workspace makes the assistant more useful for your specific project. The journalist toolkit team can share skills across team members — one person writes the source-finder skill, another writes the timeline generator, and everyone benefits.</aside>
 </section>
 
 
@@ -959,7 +959,7 @@ When I ask you to prepare for an interview:
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
-  <aside class="notes">This comes from Thariq at Anthropic — the person who built Claude Code. After cataloging hundreds of internal skills, they found 9 recurring categories. The best skills fit cleanly into one; the confusing ones straddle several. For journalism: your daily digest is a "Business Process Automation" skill (category 4). Your interview prep is a "Runbook" (category 8) — symptom is "I have an interview," investigation is research, report is the question list. Think about which categories match YOUR workflows. The homework: pick a category and design a skill for it.</aside>
+  <aside class="notes">This comes from Thariq at Anthropic — the person who built Claude Code. After cataloging hundreds of internal skills, they found 9 recurring categories. The best skills fit cleanly into one; the confusing ones straddle several. For journalism: your daily digest is a "Business Process Automation" skill (category 4). Your interview prep is a "Runbook" (category 8) — symptom is "I have an interview," investigation is research, report is the question list. MAP YOUR CAPSTONE: which category does your first skill fall into? The URL-to-video pipeline is "Business Process Automation." The headline refiner is "Product Verification" — it checks output quality. The political calendar is "Data Fetching &amp; Analysis." The Slack emoji game is a "Runbook" with a playful twist. Knowing the category helps you write a better skill because you can follow that category's pattern.</aside>
 </section>
 
 
@@ -1035,6 +1035,7 @@ When I ask you to prepare for an interview:
     </div>
   </div>
   <div class="slide-footer">FLORENT DAUDENS</div>
+  <aside class="notes">The capstone connection is the through-line for homework. "Use it for real work" means use it for your capstone. "Improve your AGENTS.md" means describe your capstone project in it. "Think about your capstone" — by April 9 you should have your AGENTS.md describing the project, at least one skill drafted, and context files with whatever data or references your project needs. Class 6 is where we build with code, connect to external APIs via MCP, and start shipping. Come prepared with a clear picture of what your project needs from an agent.</aside>
 </section>