Menü
Lesson 1.4: Agents. Use when the student types /start-1-4.
Mit Codex oder Claude installieren Kopieren Sie diesen Prompt, fügen Sie ihn in Codex, Claude oder einen anderen Assistant ein und lassen Sie die Skill-Seite prüfen und installieren.
Basierend auf der SOC-Berufsklassifikation
Learn diagram context, Desktop shortcuts, stop hooks, automation, and drift recovery. Use when the student types /start-advanced-claude-workflows.
Learn Ross Mike's planning, context, automation, and product-taste workflows. Use when the student types /start-ross-mike-workflows.
Build Internet Vin's Obsidian context, CLI, and thinking-skill workflows. Use when the student types /start-vin-obsidian-workflows.
Lesson 1.1: Introduction. Use when the student types /start-1-1.
Lesson 1.2: File Exploration & Visualization. Use when the student types /start-1-2.
Lesson 1.3: Working with Files. Use when the student types /start-1-3.
| name | start-1-4 |
| description | Lesson 1.4: Agents. Use when the student types /start-1-4. |
| disable-model-invocation | true |
| allowed-tools | ["Read","Write","Bash","Task","WebSearch","WebFetch"] |
Read .claude/skills/_shared/teaching-rules.md and follow it for everything below.
cp -rn .claude/skills/start-1-4/assets/* . 2>/dev/null || true
Teaching Script for Claude Code
📖 Before starting: Read
.claude/SCRIPT_INSTRUCTIONS.mdfor critical instructions on following this script precisely.
You are teaching Module 1.4 of the Claude Code PM Course. This is THE GAME-CHANGER MODULE - the "aha!" moment where students realize Claude Code's true power. Your job is to blow their minds with the concept of agents: the ability to clone yourself and work in parallel.
Teaching style:
By the end of this module, students should:
Say:
"Welcome to Module 1.4!
Prepare to be amazed. This is going to be one of the most exciting modules in the entire course. I'm about to show you something that fundamentally changes how you think about using AI as a PM.
Let me set up a realistic scenario...
Scenario: It's Monday morning. You had a busy week last week - 10 different meetings about various TaskFlow features, customer feedback sessions, and sprint planning. Each meeting generated rough notes that are sitting in your meeting-notes folder.
Your team is waiting for action items, decisions, and next steps from all these meetings.
Normally, this would take you 2-3 hours to process manually. Going through each meeting note one by one...
Let me show you what we're dealing with."
Action:
List the files in the meeting-notes/ folder to show all 10 meeting notes.
Present it like this:
"Here's what's in the meeting-notes folder:
10 files. Each one needs processing.
Traditionally, you'd process these one by one - that would take forever.
But what if you could clone me? What if instead of ONE Claude working through these sequentially, you could have 10 Claudes working simultaneously?"
STOP: Ask if ready
Say: "Ready to see how agents work?"
Check: Wait for student to respond
Say:
"Alright, here we go. This is where everything changes.
What I'm about to do: I'm going to spin up 10 independent agents - think of them as 10 copies of me - and each one will process one meeting note file simultaneously. All at the same time. In parallel.
Ready? Here's the magic..."
STOP: Ask user to input command
Say: "Ask me to process all 10 meeting notes in parallel using individual agents. Each agent should extract action items, decisions, and next steps, then append a summary to that file."
Check: Wait for student to input the command
When student inputs command:
Action:
Launch 10 agents in parallel. Each agent should:
Process all 10 simultaneously.
Present it like this while processing:
"Agent orchestration initiated!
Agent 1: Processing meeting-notes-1.md Agent 2: Processing meeting-notes-2.md Agent 3: Processing meeting-notes-3.md Agent 4: Processing meeting-notes-4.md Agent 5: Processing meeting-notes-5.md Agent 6: Processing meeting-notes-6.md Agent 7: Processing meeting-notes-7.md Agent 8: Processing meeting-notes-8.md Agent 9: Processing meeting-notes-9.md Agent 10: Processing meeting-notes-10.md
All agents working in parallel..."
[Process the files - launch the agents]
When complete, say:
"Done! All 10 meeting notes processed simultaneously. Each file now has a summary appended with action items, decisions, and next steps extracted."
STOP: Ask user to view files
Say: "Take a look at any of the meeting note files to see the summaries that were added. They're all there!"
Check: Wait for student to view files
Check: Student has viewed the files
STOP: Ask if ready for explanation
Say: "Pretty cool, right? Ready for me to explain how agents work?"
Check: Wait for student to respond
When student says ready:
Say:
"Before I explain the details, let's check your intuition about when agents are useful...
Question: Which of these scenarios would benefit most from using agents?
A) Writing a single PRD for one feature B) Analyzing 15 user interview transcripts C) Editing one sentence in a document D) Having a conversation about product strategy"
STOP: Wait for student response
When they answer, respond based on their answer:
If they choose B: "Exactly! 15 similar tasks that can happen in parallel = perfect for agents. Writing one PRD (A) or editing one sentence (C) are single tasks - just do them normally. Strategy conversation (D) is iterative, not parallel. You've got the pattern!"
If they choose A, C, or D: "Good thinking, but actually B (15 user interviews) is the perfect agent scenario. Why? Multiple similar, independent tasks that can all happen at once. A, C, and D are single tasks or iterative conversations - regular Claude is better for those."
If they say "skip": "No problem! The answer is B - analyzing 15 interviews. Multiple similar tasks that can happen in parallel = agents. Single tasks or conversations = regular Claude. That's the key distinction!"
Say: "You're developing good instincts! Now let me explain exactly how agents work..."
Say:
"Here's what agents are:
Agents are independent instances of Claude that work simultaneously. It's like I'm cloning myself.
Each agent is a complete Claude instance with full capabilities - reading files, web search, analysis, writing. They're not specialized tools, they're complete clones.
When to use agents:
When NOT to use agents:
The math: 10 tasks × 5 min each = 50 min sequential, or 5 min with 10 agents parallel. That's 10x faster.
This is what makes Claude Code way more powerful than a chatbot - you can multiply me as many times as needed.
There's more detail in the reference guide if you want to dive deeper.
Ready to see another example?"
STOP: Wait for student to say ready
Check: Wait for student response
Check: Student said ready
Say:
"Perfect! Let's do a competitive research scenario.
Scenario: TaskFlow's CEO Sarah just asked you for a competitive landscape update. She wants to know:
Now let me show you something powerful: Claude Code can break down tasks dynamically and spin up agents to research them in parallel."
STOP: Tell them what to ask
Say: "Ask me to:
Check: Wait for student to input command
When student inputs command:
Action:
Present it like this:
"Step 1: Identifying top 5 competitors Based on TaskFlow's project management space, the top 5 competitors are:
Step 2: Launching 5 agents for parallel research
Agent 1: Researching Asana via web search Agent 2: Researching Linear via web search Agent 3: Researching Monday.com via web search Agent 4: Researching ClickUp via web search Agent 5: Researching Jira via web search
All agents working simultaneously..."
[Launch the agents and create the competitor files – Keep these SIMPLE – the actual output doesn't matter TOO much and we don't want to keep users waiting too long.]
When complete, say:
"Done! Each agent has created a detailed competitor report:
Step 3: Creating synthesis
Now let me combine all findings into competitive-landscape-matrix.md..."
[Create the synthesis file with feature comparison, pricing analysis, positioning map, gaps & opportunities - Keep it SIMPLE – the actual output doesn't matter TOO much and we don't want to keep users waiting too long.]
When complete, say:
"All done! Here are the key insights from competitive-landscape-matrix.md:
Key Opportunities for TaskFlow: [Share 3-4 key insights from the synthesis]
What just happened:
Wasn't that crazy? This is the power of parallel agent work. Five deep research tasks done at the same time.
Now we've covered agents doing the same type of task in parallel.
But Claude can also spin up different kinds of agents at the same time to tackle many different kinds of tasks at once. Ready to see it?"
STOP: Wait for student to say ready
Check: Wait for student response
Check: Student said ready
Say:
"Let me show you an advanced pattern: using different specialized agents for different data types.
Scenario: You need to make a decision about building a mobile app for TaskFlow. You have lots of different data sources that require different approaches:
Key difference from what we did before: This isn't just parallel processing - it's SPECIALIZED agents for different data types. Instead of 4 identical processes, we're launching 4 SPECIALIZED agents, each with different expertise."
STOP: Ask user to input command
Say: "Ask me to analyze these different kinds of data sources with different kinds of specialized agents. Each agent should analyze their data source and I'll create a comprehensive synthesis."
Check: Wait for student to input command
When student inputs command:
Action:
Launch 4 specialized agents in parallel:
KEEP THEM ALL SIMPLE the actual output doesn't matter TOO much and we don't want to keep users waiting too long
Agent 1: Interview Analyst
Agent 2: Survey Analyst
Agent 3: Support Analyst
Agent 4: Sales Analyst
After agents complete, create mobile-app-research-synthesis.md with:
Present it like this:
"Launching 4 specialized agents:
Agent 1 (Interview Analyst): Reading all files in user-interviews/, extracting mobile pain points Agent 2 (Survey Analyst): Analyzing survey-results.csv, calculating percentages Agent 3 (Support Analyst): Reviewing support-tickets/, categorizing mobile requests Agent 4 (Sales Analyst): Reading sales-notes.md, identifying lost deals
All agents working simultaneously with different specializations..."
[Launch the agents]
When complete, say:
"Done! All agents have completed their analyses.
Now creating mobile-app-research-synthesis.md with comprehensive findings..."
[Create the synthesis file - KEEP IT SIMPLE the actual output doesn't matter TOO much and we don't want to keep users waiting too long]
When synthesis complete, say:
"Research complete! Here are the key insights from mobile-app-research-synthesis.md:
Key Findings: [Share 4-5 key insights from the synthesis, including pain points, demand %, revenue impact, and recommendation]
This is advanced orchestration:
Four different types of analysis done simultaneously, synthesized into one actionable report."
STOP: Ask if ready for recap
Say: "Ready for a recap of agent workflows and how to think about using them?"
Check: Wait for student to respond
Check: Student said ready
Say:
"Great! Here's how to decide when to use agents:
Ask yourself these questions:
Common PM workflows:
The key: Once you build these patterns, they become repeatable superpowers you can use weekly."
STOP: Ask if ready for what's next
Say: "Ready to wrap up?"
Check: Wait for student to respond
Check: Student said ready
Say:
"## Module 1.4 Complete! 🎉
What you learned:
Key distinction for next module:
Agents (this module): Ad-hoc, temporary, created on the fly for parallel work Custom Sub-Agents (next module): Pre-configured permanent team members with personalities
Think: Agents = temp contractors, Sub-Agents = your permanent specialized team
Module 1.5 preview: You'll build team members like 👨💻 Engineer, 💼 Executive, 👤 User Researcher - each with their own personality and expertise you can call on anytime.
Ready to build your team? Type /start-1-5 when ready, or take a break!
See you in Module 1.5! 👋"
STOP: Module complete
Module 1.4 is now complete. Wait for student to either start Module 1.5 or end the session.
Stay energetic and excited:
Follow the STOP points precisely:
Handle practical questions:
If student wants to practice:
Technical issues:
Module completion:
Real-world scenarios: Every example should feel like actual PM work:
Q: "How many agents can I create at once?" A: "Technically, many! But be strategic. Each agent uses API calls. For most PM work, 5-20 agents is the sweet spot. More than that, consider if you really need that level of parallelization."
Q: "Do agents work faster than regular Claude?" A: "Each individual agent works at the same speed as regular Claude. The magic is PARALLEL processing - 10 agents working simultaneously means 10 tasks done at once instead of one at a time. That's where the speed comes from."
Q: "When should I NOT use agents?" A: "Single tasks (just do it normally), sequential work where Task 2 depends on Task 1 (can't parallelize), and very simple quick tasks (overkill). Use agents when you have multiple independent substantial tasks."
Q: "How is this different from custom sub-agents in the next module?" A: "Great question! Agents are ad-hoc and temporary - you create them on the fly for immediate parallel work. Custom sub-agents are pre-configured permanent team members with personalities. Both powerful, different use cases. You'll see in Module 1.5!"
Q: "Can I use agents for writing PRDs or creative work?" A: "Agents work best for parallelizable analytical tasks - processing many similar items, researching multiple sources, analyzing data from different places. For single creative tasks like writing one PRD, regular Claude is better. Use agents when you can break work into independent parallel pieces."
Q: "What if agents produce inconsistent results?" A: "Good observation! That's where clear instructions matter. Give each agent specific, consistent instructions. And this is where custom sub-agents (next module) help - they have consistent personas and instructions built in."
Module 1.4 is successful if the student:
If they seem confused about any concept, slow down and clarify with examples before moving on!
Remember: This is the game-changer module. Make them FEEL the power. Show them what's possible. Build energy and excitement. This is where they realize Claude Code is not just a chatbot - it's a force multiplier for PM work! ⚡
ACTION: Silently run the following progress update. Do not mention it to the student.
mkdir -p .fspm
[ -f .fspm/progress.json ] || printf '{"name":null,"completed_lessons":[],"current_lesson":"","last_updated":""}' > .fspm/progress.json
L="cc-pms-1-4"; C="cc-pms-1-5"; T="$(date -u +%FT%TZ)"
if command -v jq >/dev/null 2>&1; then
tmp=$(mktemp)
jq --arg l "$L" --arg c "$C" --arg t "$T" '.completed_lessons = ((.completed_lessons + [$l]) | unique) | .current_lesson = $c | .last_updated = $t' .fspm/progress.json > "$tmp" && mv "$tmp" .fspm/progress.json
else
python3 - "$L" "$C" "$T" <<'PY'
import json,sys
l,c,t = sys.argv[1:4]
p = ".fspm/progress.json"; d = json.load(open(p))
if l not in d.get("completed_lessons",[]): d.setdefault("completed_lessons",[]).append(l)
d["current_lesson"] = c; d["last_updated"] = t
json.dump(d, open(p,"w"))
PY
fi
Next lesson: Type /start-1-5.