Framework
Follow along with the written framework below as you watch the curated video
What You'll Learn
- When to continue a session vs. start fresh
- How to use
/recapeffectively (not just "save and quit") - What to do when Claude makes a mistake
- Real workflow examples (a founder's day with Claude)
- Multi-turn conversations (build → test → iterate)
Why This Matters (2 min read)
The Scenario:
You've been working with Claude for 3 hours. You've:
- Generated 5 product specs
- Analyzed 10 customer interviews
- Written landing page copy
- Created an email sequence
Now you want to switch gears and work on competitive research.
Question: Should you:
A) Keep going in the same session?
B) /recap and start fresh?
Most founders pick A. Wrong choice.
What Happens Next:
- Context is at 85% (very full)
- Claude starts "forgetting" earlier instructions
- Responses get slower
- Quality declines
- You're frustrated
You should have picked B (/recap at 75%, start fresh for new task).
This lesson teaches you:
- When to continue vs. reset
- How to maintain quality across long work sessions
- What to do when things go wrong
When to Continue vs. Start Fresh
Continue Session When:
✅ Context under 50% — Plenty of room to keep going
✅ Working on the SAME task — "Refine this product spec" (iterating)
✅ Building incrementally — "Now add a pricing section" (expanding previous work)
✅ Quick follow-ups — "Make the tone more casual"
Example:
textSession (context: 20%): You: "Write a product spec for feature X" Claude: [generates spec] You: "Great! Now add a section on success metrics" Claude: [adds section] You: "Perfect. Add user stories too." Claude: [adds stories] Context now: 35% — Still healthy, keep going!
Start Fresh When:
❌ Context hits 75-80% — Quality starts declining
❌ Switching to a NEW task — Product spec done, now doing competitive research
❌ Claude seems "confused" — Forgetting earlier instructions, contradicting itself
❌ After 2+ hours — Natural break point
❌ Major mistake — Claude misunderstood entirely, need to reset
Example:
textSession (context: 78%): You: "Now let's switch to competitive analysis" BETTER: You: "/recap" Claude: [saves session summary] You: [Start NEW session] You: "Read my recap and now let's do competitive analysis" Context now: 5% — Fresh start, high quality!
How to Use /recap Effectively
What /recap Does
- Summarizes the session — Key decisions, outputs created, next steps
- Saves to a file — You can reference it later
- Resets context — Starts a fresh session (0% context)
- Preserves continuity — You don't lose work, just compress it
Think of it like:
- Saving a video game checkpoint
- Closing a browser tab (to free up RAM)
- Taking meeting notes before ending the meeting
The /recap Workflow
Step 1: Run the command
You: /recap
Step 2: Claude generates summary
textClaude: ## Session Recap (2026-02-08, 3:45 PM) ### What We Accomplished - Generated product spec for "Task Automation" feature - Saved to: specs/task-automation-spec.md - Analyzed 10 customer interviews - Top 3 pain points identified - Saved to: analysis/customer-insights-2026-02-08.md - Wrote landing page hero copy (3 iterations) - Final version saved to: marketing/landing-hero-v3.md ### Key Decisions - Target audience: Remote teams, 10-50 people - Pricing validated: $49/user/month - Launch timeline: March 2026 ### Next Steps 1. Competitive analysis (10 competitors) 2. Feature prioritization (RICE scoring) 3. Email sequence draft (onboarding flow) Recap saved to: session-recaps/2026-02-08-product-validation.md
Step 3: Start fresh session
- Context resets to 0%
- You can now start new work with full context budget
Step 4: Continue where you left off (optional)
text[NEW SESSION] You: "Claude, read session-recaps/2026-02-08-product-validation.md and continue with competitive analysis" Claude: [reads recap, understands context] "Got it! I see we identified 3 pain points and validated pricing. Now let's analyze competitors..."
When to /recap (The Rules)
| Context Level | Action | Why |
|---------------|--------|-----|
| 0-50% | Keep going | Plenty of room |
| 50-70% | Consider recapping if switching tasks | Optional |
| 70-80% | /recap soon | Quality starts declining |
| 80%+ | /recap NOW | Risk of cutoff, poor quality |
| 100% | Too late, must recap | Can't continue |
What to Do When Claude Makes a Mistake
Claude is powerful, but not perfect. It WILL make mistakes.
Common Mistakes:
- Misunderstood your prompt — Generated wrong thing
- Overwrote the wrong file — Edited footer instead of header
- Generated broken output — Missing section, incorrect format
- Contradicted earlier instructions — "Forgot" what you said 10 prompts ago
Recovery Strategies
Mistake 1: Claude Misunderstood (Wrong Output)
What happened:
textYou: "Write a product spec" Claude: [writes marketing copy instead]
How to fix:
textYou: "That's not what I asked for. Let me clarify: I need a PRODUCT SPEC (technical requirements doc), not marketing copy. Format: ## Problem ## Solution ## Features ## Success Metrics Try again."
Why it works: You're clarifying, not blaming. Claude recalibrates.
Mistake 2: Claude Edited Wrong Thing
What happened:
textYou: "Update the hero section" Claude: [edits footer instead]
How to fix:
textYou: "Undo that. You edited the footer, not the hero. The HERO is the first section: [paste the actual hero code block] Edit THIS section, not the footer."
Why it works: You're being specific about what to edit. No ambiguity.
Mistake 3: Broken Output (Missing Sections)
What happened:
textYou: "Generate a 5-section spec" Claude: [only generates 3 sections]
How to fix:
textYou: "This is incomplete. You only included 3 sections. Add the missing 2: - Section 4: User Stories - Section 5: Success Metrics"
Why it works: Point out what's missing, Claude fills the gap.
Mistake 4: Claude "Forgot" Earlier Context
What happened:
text[20 prompts ago] You: "Use casual tone" [Now] Claude: [writes in formal, corporate tone]
How to fix (Option A): Remind Claude
You: "Remember, I asked for CASUAL tone earlier. Rewrite this in casual tone like: [show example]"
How to fix (Option B): /recap and start fresh
- If context is high (75%+), Claude is "forgetting" because context is full
- Recap, start fresh session
- Restate key preferences upfront
Multi-Turn Workflows (Build → Test → Iterate)
Most founder work isn't "one prompt → done." It's iterative.
Workflow Example: Product Spec Generation
Turn 1: Initial Draft
textYou: "Write a product spec for 'Task Automation' feature using specs/template.md" Claude: [generates first draft]
Turn 2: Review & Refine
textYou: "Good start! Two changes: 1. Add a 'Why Now' section (why this feature is important now) 2. Expand success metrics (3 metrics, not 1)" Claude: [adds sections]
Turn 3: Polish
textYou: "Almost there. Make the tone less technical (our audience is non-technical team leads)" Claude: [simplifies language]
Turn 4: Final Check
textYou: "Perfect. Save this to specs/task-automation-spec-v1.md" Claude: [saves file]
Context used: ~10% Time: 10 minutes Quality: High (because you guided Claude step-by-step)
Real Founder Workflows (A Day with Claude)
Morning: Product Validation (1 hour)
Context Budget: 0% → 40%
Session 1: Idea Validation
text9:00 AM - Context: 0% You: "I have a new feature idea: 'AI-powered task suggestions'. Help me validate it." Claude: [asks clarifying questions, generates validation doc] Context: 15% You: "Now analyze 5 competitors to see if they have this" Claude: [researches, generates competitive matrix] Context: 35% You: "Save analysis to research/ai-task-suggestions-validation.md" Claude: [saves] Context: 40% You: "/recap" (switching to customer research next)
Mid-Morning: Customer Research (1 hour)
Context Budget: 0% → 50%
Session 2: Interview Analysis
text10:30 AM - Context: 0% (fresh session) You: "Read session recap, then read 8 new customer interviews in research/interviews/" Claude: [reads recap + interviews] Context: 25% You: "Find patterns. What are the top 3 pain points?" Claude: [analyzes, generates insights doc] Context: 40% You: "Great. Now generate 3 user personas based on these interviews" Claude: [creates personas] Context: 50% You: "/recap" (lunch break)
Afternoon: Content Creation (2 hours)
Context Budget: 0% → 60%
Session 3: Landing Page + Email Sequence
text1:00 PM - Context: 0% You: "Read research/ai-task-suggestions-validation.md, then write landing page hero copy" Claude: [reads validation, writes copy] Context: 20% You: "Good. Iterate 3 times (make it punchier)" Claude: [refines] Context: 30% You: "Now write a 5-email onboarding sequence" Claude: [generates sequence] Context: 55% You: "Save hero to marketing/landing-hero.md and emails to marketing/onboarding-sequence.md" Claude: [saves] Context: 60% You: "/recap" (end of work day)
Evening: Prepare for Tomorrow (15 min)
Review Today's Outputs
text6:00 PM You: "Summarize what we accomplished today across all 3 sessions" Claude: [reads all 3 recaps, generates daily summary] Daily Summary: - Validated new feature idea (AI task suggestions) - Analyzed 8 customer interviews → 3 personas - Created landing page hero copy (3 iterations) - Wrote 5-email onboarding sequence Tomorrow's priorities: - Build product spec for AI task suggestions feature - Create competitive positioning doc - Draft investor update email
Total Context Used Today: ~150% (but spread across 3 sessions, so quality stayed high)
Pro Tips for Session Management
Tip 1: Name Your Sessions (Use Descriptive Prompts)
❌ Bad:
textYou: "Help me" [Session saved as "Untitled Conversation"]
✅ Good:
textYou: "Let's work on product validation for the AI task suggestions feature" [Session saved as "Product Validation - AI Task Suggestions"]
Why: Easy to find in history when you need to resume.
Tip 2: Recap at Natural Break Points
- End of a major task
- Lunch break
- Context at 75%
- Switching topics
Don't: Recap mid-task (you'll lose momentum)
Tip 3: Read Previous Recaps to Resume
Scenario: It's been 3 days since you worked on this project.
Don't:
textYou: "Where did we leave off?" Claude: "I don't have access to previous sessions"
Do:
textYou: "Read session-recaps/2026-02-05-product-validation.md and continue where we left off" Claude: [reads recap, understands full context] "Got it! We validated the feature and analyzed competitors. Next step was building the product spec. Let's do that now."
Tip 4: Use CLAUDE.md to Reduce Repeated Context
If you're ALWAYS saying "I'm a non-technical founder..." every session:
Put it in CLAUDE.md (permanent memory)
- Never repeat again
- Saves context
- Faster sessions
Try This Now (Exercise — 10 Minutes)
Step 1: Start a Short Session (5 min)
- Ask Claude to help with a small task (write 200-word product pitch)
- Iterate 2-3 times ("Make it punchier", "Add a CTA")
- Check context usage (
/usage)
Step 2: Practice /recap (3 min)
- Run
/recap - Read the recap Claude generates
- Note where it's saved
Step 3: Resume in New Session (2 min)
- Start a NEW Claude Code session
- Say: "Read [recap file] and continue"
- Ask Claude: "What did we work on last session?"
- Claude should summarize accurately
What You Learned:
- How to iterate within a session
- How
/recapworks - How to resume seamlessly
Common Mistakes (And Fixes)
Mistake 1: Never Recapping
Problem: Work for 4 hours straight, context at 95%, Claude struggling.
Fix: Recap every 2 hours or at 75% context.
Mistake 2: Recapping Mid-Task
Problem: Claude is generating a 5-page spec, you run /recap after page 2.
Fix: Let Claude finish the task, THEN recap.
Mistake 3: Not Reading Recaps Later
Problem: You recap every session but never read them again. Waste of time.
Fix: When resuming work, read previous recap first. Claude needs context too.
Mistake 4: Expecting Perfect Memory
Problem: 50 prompts ago you said X, Claude "forgot."
Fix: If it's important, put it in CLAUDE.md. Or remind Claude when needed.
Downloads
📥 Session Management Cheat Sheet (PDF) — Quick reference 📥 Recap Template — Manual recap structure 📥 Daily Workflow Template — Morning/afternoon session plan 📥 Troubleshooting Guide — When things go wrong
Module 0 Complete! 🎉
You've finished Claude Code Bootcamp!
You now know: ✅ How to organize projects for AI (Lesson 0.1) ✅ How to manage context without running out (Lesson 0.2) ✅ How to write effective prompts (Lesson 0.3) ✅ How to use CLAUDE.md for persistent memory (Lesson 0.4) ✅ How to manage sessions like a pro (Lesson 0.5)
What's Next: Founder Track (Modules 1-5)
Now that you KNOW how to use Claude Code, let's apply it to REAL founder work:
Module 1: AI for Product Development (4 lessons)
- Validating ideas with AI
- Competitive research automation
- Feature spec generation
- User story writing
Module 2: AI for Customer Research (4 lessons)
- Interview question generation
- Transcript synthesis
- Pattern finding in feedback
- Building user personas
Module 3: AI for Workflow Automation (4 lessons)
- When to automate vs. stay manual
- No-code tools (Zapier, Make, n8n)
- Building your first workflow
- AI for meeting notes
Module 4: AI for Content & Marketing (3 lessons)
- Landing page copy
- Social media automation
- Email campaigns
Module 5: AI for Strategic Thinking (3 lessons)
- Market analysis
- Competitive intelligence
- OKR planning
→ Start Module 1: AI for Product Development
Key Takeaways
✅ Continue when context < 50% and working on same task
✅ Recap when context hits 75-80% or switching tasks
✅ Use /recap at natural break points (not mid-task)
✅ Read previous recaps when resuming work
✅ Fix mistakes by clarifying (not blaming)
✅ Iterate in multi-turn workflows (build → test → refine)
Sources & Attribution:
This lesson curates content from:
- Anthropic: Context Management
- Claude Code Best Practices Playbook
- Claude Code Official Docs: Checkpointing
Additional insights and founder workflow examples by NerdSmith.
Version: 1.0 (2026-02-08) Module: 0 — Claude Code Bootcamp Lesson: 5 of 5 (Complete!)