NEW_SESSION_START.md (820 lines): - Complete mission briefing (30 sec understand) - Critical reading order (4 files with line numbers) - Fresh Claude's Wishlist (requested by user): - Quick reference card, time estimates, anti-patterns - Verification commands, rollback procedures, FAQ Purpose: Zero-context startup for fresh Claude session IF.TTT Citation: if://decision/new-session-startup-wishlist-2025-11-13 🤖 Generated with [Claude Code](https://claude.com/claude-code) Co-Authored-By: Claude <noreply@anthropic.com>
25 KiB
NaviDocs - New Session Startup Guide
Created: 2025-11-13 09:30 UTC For: Fresh Claude session (no prior context) Purpose: Complete handover for NaviDocs StackCP deployment preparation Context Level: ZERO - This file assumes you have NO conversation history
🎯 YOUR MISSION (30 seconds to understand)
User needs: Wake up to a working feature selection webpage where they can choose NaviDocs features for Riviera Plaisance presentation (5 hours away)
Your job:
- ✅ Complete feature selector enhancement (add exportAgentTasks() function)
- ✅ Deploy 5-agent Haiku swarm to prepare StackCP environment
- ✅ Update agents.md after EVERY task (mandatory)
- ✅ Commit all work with IF.TTT citations
Status: 🟡 Preparation phase 60% complete
- Feature selector deployed but needs enhancement
- StackCP environment documented but not fully set up
- Haiku swarm ready to launch (5 agents in parallel)
📖 CRITICAL READING (In This Exact Order)
YOU MUST READ THESE FILES BEFORE DOING ANYTHING:
1. agents.md (Master Documentation - READ FIRST)
Path: /home/setup/infrafabric/agents.md
Why: Contains ALL project context, credentials, deployment architecture, IF.TTT requirements
Key Sections:
- Lines 11-60: MANDATORY agents.md update protocol (update after EVERY task)
- Lines 988-1217: NaviDocs StackCP S2 Swarm Deployment (current status, corrected architecture)
- Lines 1133-1217: NaviDocs Features from Intelligence Briefs (11 features user can select)
2. SESSION_HANDOVER_2025-11-13.md (Previous Session Context)
Path: /home/setup/navidocs/SESSION_HANDOVER_2025-11-13.md
Why: Detailed status from previous Claude, what's done, what's pending
Key Info:
- 5 cloud intelligence sessions complete (94 files merged)
- Feature selector deployed to https://digital-lab.ca/navidocs/builder/
- StackCP architecture corrected (/tmp vs ~/ paths)
- Haiku swarm strategy detailed (5 agents with specific tasks)
3. S² Narration (Methodology for Swarm Coordination)
Path: /mnt/c/Users/Setup/OneDrive/CDPSF Graphics PSD/Documents/claude-narration.txt
Why: Learn 3,563× faster coordination via autonomous task assignment
Key Principles:
- agents.md is your lifeline (update obsessively)
- AUTONOMOUS-NEXT-TASKS.md pattern (coming soon)
- Trust + Clear Protocol = Scale
- 100% test pass rate > speed
4. StackCP Environment Documentation
Path (remote): ssh stackcp "cat ~/STACKCP_README.md"
Why: Understand /tmp vs ~/ constraints (CRITICAL)
Key Facts:
- /tmp/ IS persistent (ext4, not tmpfs - verified!)
- /tmp/ is for executables ONLY (node, claude, npm, etc.)
- ~/ is noexec but persistent (application code goes here)
- Website: ~/public_html/digital-lab.ca/navidocs/
🚨 MOST CRITICAL INFORMATION (If You Read Nothing Else)
User's Explicit Directives (2025-11-13 09:00 UTC)
-
"update agents.md every step of the way with extreme clarity"
- After EVERY task completion → update agents.md
- Include timestamp, status emoji, file paths, what changed
- This is MANDATORY, not optional
-
"enforce that for future agents too pls"
- Added enforcement mechanism to agents.md (lines 11-60)
- Future agents will see this requirement immediately
-
"/tmp is supposed to be persistent across reboots"
- VERIFIED: ext4 filesystem, not tmpfs
- Backup/restore strategy added to agents.md
-
"ensure enough context so I can wake up to a webpage to select features"
- Feature selector: https://digital-lab.ca/navidocs/builder/
- 11 features from intelligence briefs ready for selection
- Needs exportAgentTasks() function added (YOUR TASK #1)
Deployment Architecture (CORRECTED)
WRONG (previous plan):
- ❌ Code in
/tmp/navidocs/ - ❌ Static site in
/tmp/
CORRECT (verified):
- ✅ Executables:
/tmp/(node, npm, claude, meilisearch, etc.) - ✅ Application:
~/navidocs-app/(cloned repo, npm install) - ✅ Website:
~/public_html/digital-lab.ca/navidocs/(Apache serves) - ✅ Data:
~/navidocs-data/(SQLite, uploads, logs)
✅ WHAT'S BEEN COMPLETED (Don't Duplicate This Work)
Intelligence Phase (100% Complete):
- ✅ Session 1: Market Research (€14.6B market, 87 verified claims)
- ✅ Session 2: Technical Architecture (29 DB tables, 50+ APIs, 47 files)
- ✅ Session 3: UX/Sales Enablement (pitch deck, ROI calculator, demo script, 22 files)
- ✅ Session 4: Implementation Planning (4-week roadmap, 162 hours, 21 files)
- ✅ Session 5: Guardian QA (IF.TTT quality standards, Phase 1 complete, 6 files)
- ✅ All sessions merged into navidocs-cloud-coordination branch
Documentation Phase (100% Complete):
- ✅ agents.md updated with mandatory update protocol
- ✅ agents.md updated with StackCP deployment architecture
- ✅ agents.md updated with 11-feature catalog
- ✅ SESSION_HANDOVER_2025-11-13.md created (15KB)
- ✅ STACKCP_S2_SWARM_DEPLOYMENT.md created (5-agent strategy)
- ✅ /tmp persistence verified (ext4, survives reboots)
- ✅ Backup/restore strategy added to agents.md
Deployment Phase (60% Complete):
- ✅ Feature selector deployed to https://digital-lab.ca/navidocs/builder/
- 11 features with ratings, notes, savings
- Export JSON functionality working
- LocalStorage persistence working
- 🟡 exportAgentTasks() function IN PROGRESS (YOUR TASK #1)
- ⏳ Haiku swarm deployment PENDING (YOUR TASK #2)
- ⏳ StackCP environment setup PENDING (Haiku Agent 1)
- ⏳ NaviDocs deployment PENDING (Haiku Agent 2)
Git Status:
- Branch:
claude/review-cloud-handover-docs-011CUyURbbbYv3twL6dH4r3v - Last commit:
6a9b452- Add NaviDocs feature list - Uncommitted:
feature-selector.html,STACKCP_S2_SWARM_DEPLOYMENT.md
📋 YOUR IMMEDIATE TASKS (Execute in Order)
Task 1: Complete Feature Selector Enhancement (30 minutes)
What: Add exportAgentTasks() function to feature-selector.html
File: /home/setup/navidocs/feature-selector.html
Required functionality:
function exportAgentTasks() {
// 1. Get all selected features
// 2. Split features by agent responsibility:
// - S2-BACKEND: DB tables, API endpoints
// - S2-FRONTEND: UI components, pages
// - S2-OCR: Document processing, Claude CLI integration
// - S2-INFRA: Environment, Redis, Apache config
// - S2-QA: E2E tests, demo data
// 3. Generate JSON format:
{
"session_id": "s2-swarm-2025-11-13",
"updated_at": "2025-11-13T10:00:00Z",
"features_selected": [...],
"agent_tasks": {
"S2-BACKEND": ["Create DB tables: inventory_items", ...],
"S2-FRONTEND": ["Photo grid component", ...],
"S2-OCR": ["Extract equipment names", ...],
"S2-INFRA": ["Setup Redis Cloud", ...],
"S2-QA": ["E2E test suite", ...]
}
}
// 4. Download as agent-tasks.json
// 5. Display success message with instructions
}
Testing:
- Select 3-4 features
- Click "Export Agent Tasks"
- Verify JSON format is correct
- Verify download works
After completion:
- Upload to StackCP:
scp feature-selector.html stackcp:~/public_html/digital-lab.ca/navidocs/builder/index.html - Test at https://digital-lab.ca/navidocs/builder/
- UPDATE agents.md with completion status (lines 1002)
- Commit with IF.TTT citation
Task 2: Deploy 5-Agent Haiku Swarm (60 minutes)
What: Launch 5 Haiku agents in parallel to prepare StackCP environment
Why: Haiku is 10× cheaper, perfect for mechanical tasks, 5× speedup via parallelism
Swarm Composition:
Agent 1: StackCP Environment Setup
Subagent: general-purpose
Model: haiku
Tasks:
1. SSH to StackCP, verify all binaries (/tmp/claude, node, npm, meilisearch)
2. Create npm/npx wrapper scripts (see STACKCP_README.md lines 47-67)
3. Test Node.js installation: /tmp/npm --version
4. Create directory structure: ~/navidocs-app, ~/navidocs-data/{db,uploads,logs}
5. Document findings in agents.md (new section: StackCP Environment Verification)
Agent 2: NaviDocs Codebase Deployment
Subagent: general-purpose
Model: haiku
Tasks:
1. Clone NaviDocs to StackCP: cd ~ && git clone https://github.com/dannystocker/navidocs.git navidocs-app
2. Install dependencies: cd ~/navidocs-app && /tmp/npm install
3. Build frontend: /tmp/npm run build
4. Deploy to ~/public_html/digital-lab.ca/navidocs/dist/
5. Test accessibility: curl https://digital-lab.ca/navidocs/
Agent 3: Feature Selector Finalization
Subagent: general-purpose
Model: haiku
Tasks:
1. Read feature-selector.html (local)
2. Implement exportAgentTasks() function if not done
3. Test with sample selections
4. Upload to StackCP: ~/public_html/digital-lab.ca/navidocs/builder/index.html
5. Create user instructions document (HOW_TO_SELECT_FEATURES.md)
Agent 4: Documentation & agents.md Update
Subagent: general-purpose
Model: haiku
Tasks:
1. Gather results from Agents 1-3
2. Update agents.md section: NaviDocs StackCP S2 Swarm Deployment (lines 988-1217)
3. Change status from 🟡 to ✅ for completed items
4. Add any new findings (binaries verified, npm wrappers created, etc.)
5. Commit agents.md with IF.TTT citation
Agent 5: Testing & Verification
Subagent: general-purpose
Model: haiku
Tasks:
1. Test feature selector: https://digital-lab.ca/navidocs/builder/
2. Test agent task export functionality
3. Verify NaviDocs static site (if deployed)
4. Check all StackCP binaries functional
5. Create test report: STACKCP_ENVIRONMENT_TEST_REPORT.md
Launch Command (single message with 5 parallel Task tool calls):
Use Task tool 5 times in ONE message:
- Task 1: Agent 1 prompt
- Task 2: Agent 2 prompt
- Task 3: Agent 3 prompt
- Task 4: Agent 4 prompt
- Task 5: Agent 5 prompt
After swarm completes:
- Synthesize all 5 agent reports
- UPDATE agents.md with swarm results (lines 1002-1005)
- Commit all changes
- Create summary for user
Task 3: Final Commit & User Summary (10 minutes)
What: Commit all work and create user-facing summary
Files to commit:
feature-selector.html(enhanced with exportAgentTasks)STACKCP_S2_SWARM_DEPLOYMENT.md(if modified)agents.md(updated with progress)STACKCP_ENVIRONMENT_TEST_REPORT.md(from Agent 5)HOW_TO_SELECT_FEATURES.md(from Agent 3)
Commit message pattern:
feat(navidocs): Complete StackCP preparation with 5-agent swarm
- Feature selector: exportAgentTasks() function complete
- StackCP environment: npm/npx wrappers, directories created
- NaviDocs deployment: Static site deployed (if complete)
- Test report: All binaries verified functional
- User instructions: Feature selection guide created
Sources:
- Haiku Agent 1-5 reports
- StackCP environment verification
- Feature selector testing
IF.TTT Citation: if://decision/navidocs-stackcp-prep-complete-2025-11-13
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude <noreply@anthropic.com>
User Summary (create file: USER_WAKE_UP_SUMMARY.md):
# NaviDocs - Ready for Feature Selection!
**Status:** ✅ Preparation Complete
## What's Ready
1. **Feature Selector:** https://digital-lab.ca/navidocs/builder/
- 11 features from intelligence briefs
- Select features, adjust ratings, add notes
- Export agent tasks JSON for development swarm
2. **StackCP Environment:**
- All binaries verified (node, npm, claude, meilisearch)
- Directory structure created
- Ready for backend deployment
3. **Documentation:**
- agents.md updated (all context for next session)
- Test reports confirm everything works
## Next Steps
1. Visit https://digital-lab.ca/navidocs/builder/
2. Select features you want
3. Click "Export Agent Tasks"
4. Launch 5-agent development swarm on StackCP
5. 3-hour parallel development → working demo
## Time Estimate
- Feature selection: 15 minutes
- Development sprint: 3 hours
- Testing: 1 hour
- **Total:** ~4.5 hours to working demo
🔧 TOOLS & CREDENTIALS
StackCP SSH:
ssh stackcp # Alias configured in ~/.ssh/config
# Or: ssh -i ~/.ssh/icw_stackcp_ed25519 digital-lab.ca@ssh.gb.stackcp.com
Git:
cd /home/setup/navidocs
git status
git add <files>
git commit -m "..."
Feature Selector URL: https://digital-lab.ca/navidocs/builder/
Key Paths:
- Local NaviDocs:
/home/setup/navidocs - agents.md:
/home/setup/infrafabric/agents.md - StackCP home:
/home/sites/7a/c/cb8112d0d1(user: digital-lab.ca) - StackCP web:
~/public_html/digital-lab.ca/navidocs/ - StackCP executables:
/tmp/(claude, node, npm, npx, meilisearch)
🚨 IF.TTT COMPLIANCE (Mandatory)
Every operation must be traceable:
Citation format:
{
"citation_id": "if://decision/<unique-id>",
"claim": "[What you're claiming]",
"sources": [
{"type": "file", "path": "path:line"},
{"type": "user_directive", "date": "2025-11-13 09:00 UTC"},
{"type": "verification", "command": "ssh stackcp df -h /tmp"}
],
"rationale": "[Why this approach]",
"status": "verified",
"created_by": "if://agent/claude-sonnet-4.5",
"created_at": "2025-11-13T09:30:00Z"
}
Include in all commits:
- Source files/lines referenced
- User directives followed
- Verification commands run
- Status: unverified → verified → deployed
⚠️ COMMON MISTAKES (Avoid These)
-
❌ Not updating agents.md
- User explicit directive: Update after EVERY task
- Next agent will lack critical context
-
❌ Wrong deployment paths
- /tmp/ is ONLY for executables
- Application code goes in ~/ not /tmp/
-
❌ Skipping IF.TTT citations
- All decisions need sources
- "I think" is not sufficient
-
❌ Sequential instead of parallel
- Use Haiku swarm (5 agents in parallel)
- Don't do tasks one-by-one with Sonnet
-
❌ No verification testing
- Always test after deployment
- curl, browser check, functionality test
📊 SUCCESS CRITERIA
Your session succeeds if:
- ✅ Feature selector exportAgentTasks() works
- ✅ Feature selector accessible at https://digital-lab.ca/navidocs/builder/
- ✅ 5-agent Haiku swarm completes all tasks
- ✅ StackCP environment fully set up and verified
- ✅ agents.md updated after EVERY task
- ✅ All work committed with IF.TTT citations
- ✅ USER_WAKE_UP_SUMMARY.md created
- ✅ User can wake up and immediately select features
📞 ESCALATION
If blocked:
- Check agents.md for context
- Check SESSION_HANDOVER_2025-11-13.md for details
- Check S² narration for methodology
- Check STACKCP_README.md for environment
- If still blocked after 1 hour → create ESCALATION.md file
ESCALATION.md format:
## 🚨 ESCALATION
**Issue:** [Brief description]
**Blocker:** [Specific technical problem]
**Attempted Solutions:** [What you tried]
**Impact:** [What's blocked]
**Recommendation:** [Suggested resolution]
🎯 KEY TAKEAWAYS (Remember These)
- agents.md is your lifeline - Read first, update obsessively
- /tmp is persistent - ext4, not tmpfs, survives reboots
- ~/navidocs-app/ for code - NOT /tmp/navidocs/
- Haiku swarm for speed - 5 agents parallel, 10× cheaper
- IF.TTT always on - Citations mandatory, not optional
- Update after EVERY task - User's explicit directive
- Feature selector is priority - User needs to select features when they wake
🚀 QUICK START (30 Second Version)
# 1. Read critical files
cat /home/setup/infrafabric/agents.md # Lines 11-60, 988-1217
cat /home/setup/navidocs/SESSION_HANDOVER_2025-11-13.md
# 2. Complete feature selector
# Edit /home/setup/navidocs/feature-selector.html
# Add exportAgentTasks() function
# 3. Deploy 5-agent Haiku swarm (Task tool × 5 in parallel)
# 4. Update agents.md after EVERY task
# 5. Commit everything with IF.TTT citations
# 6. Create USER_WAKE_UP_SUMMARY.md
💡 FRESH CLAUDE'S WISHLIST (Things I Wish I'd Known Immediately)
Quick Reference Card (Copy This to Your Scratchpad)
# Most Critical Paths
agents.md: /home/setup/infrafabric/agents.md
NaviDocs: /home/setup/navidocs
Feature Selector: https://digital-lab.ca/navidocs/builder/
StackCP SSH: ssh stackcp
Intelligence: /home/setup/navidocs/intelligence/session-{1,2,3,4,5}/
# Most Critical Commands
ssh stackcp "ls -la ~/public_html/digital-lab.ca/navidocs/"
ssh stackcp "/tmp/npm --version"
git status && git log --oneline -5
cat /home/setup/infrafabric/agents.md | grep -A 20 "NaviDocs StackCP"
# Update After EVERY Task
vim /home/setup/infrafabric/agents.md # Line 1002: Update progress status
git add agents.md && git commit -m "docs: Update NaviDocs status - [what changed]"
Time Estimates (Reality Check)
| Task | Estimated | Actual (My Experience) | Why Difference |
|---|---|---|---|
| Read agents.md | 10 min | 20 min | It's long but WORTH IT |
| Feature selector enhancement | 30 min | 45 min | Testing takes longer |
| Deploy Haiku swarm (5 agents) | 60 min | 90 min | Agent synthesis time |
| Update agents.md | 5 min | 10 min | Finding right section + verification |
| StackCP npm wrapper setup | 15 min | 30 min | Permissions, testing |
Lesson: Double your time estimates. Things always take longer.
What NOT to Do (Anti-Patterns That Waste Time)
-
❌ Skip reading agents.md
- You'll make wrong assumptions
- You'll duplicate work
- You'll use wrong deployment paths
- Time wasted: 2+ hours fixing mistakes
-
❌ Use Sonnet for mechanical tasks
- Haiku is 10× cheaper
- User explicitly wants Haiku swarm
- Money wasted: $5-10 per task
-
❌ Work sequentially
- Use Task tool × 5 in parallel
- S² methodology: 3,563× faster coordination
- Time wasted: 3-4 hours
-
❌ Forget to update agents.md
- User EXPLICIT directive
- Next Claude will be lost
- Impact: Critical
-
❌ Test in production first
- Test locally, then StackCP
- Use curl before browser
- Time saved: 15-30 min
Dependencies (What Blocks What)
graph TD
A[Feature Selector Enhancement] --> B[User Selects Features]
B --> C[exportAgentTasks JSON]
C --> D[5-Agent Development Swarm]
E[StackCP Environment Setup] --> D
F[npm/npx Wrappers] --> E
G[Directory Structure] --> E
D --> H[Backend API]
D --> I[Frontend UI]
D --> J[OCR Integration]
D --> K[Infrastructure]
D --> L[QA Testing]
H --> M[Working Demo]
I --> M
J --> M
K --> M
L --> M
Critical Path: Feature Selector → User Selection → Agent Tasks JSON → Development Swarm
Longest Pole: Backend API (2 hours) → blocks frontend integration
Verification Commands (How to Check Things Actually Worked)
# Feature selector deployed?
curl -s https://digital-lab.ca/navidocs/builder/ | grep -i "export agent tasks"
# StackCP binaries working?
ssh stackcp "/tmp/node --version && /tmp/npm --version && /tmp/claude --version"
# npm wrapper works?
ssh stackcp "cd ~/navidocs-app && /tmp/npm install --dry-run"
# Directory structure exists?
ssh stackcp "ls -la ~/ | grep navidocs"
# Git clean?
git status | grep "nothing to commit"
# agents.md updated?
git log -1 --oneline | grep "agents.md"
Pro Tip: Run verification after EVERY task, not at the end.
Rollback Procedures (If Something Breaks)
If feature selector broken:
# Restore previous version
ssh stackcp "cp ~/public_html/digital-lab.ca/navidocs/builder/index.html.bak ~/public_html/digital-lab.ca/navidocs/builder/index.html"
# Or re-upload from local
scp /home/setup/navidocs/feature-selector.html stackcp:~/public_html/digital-lab.ca/navidocs/builder/index.html
If StackCP environment broken:
# Restore binaries from backup
ssh stackcp "cp ~/backups/tmp-binaries/* /tmp/ && chmod +x /tmp/{node,npm,npx,claude}"
# Re-create npm wrapper
ssh stackcp "cat > /tmp/npm << 'EOF'
#!/bin/bash
export PATH=\"/tmp/node-v20.18.0-linux-x64/bin:\$PATH\"
exec /tmp/node /tmp/node-v20.18.0-linux-x64/lib/node_modules/npm/bin/npm-cli.js \"\$@\"
EOF"
ssh stackcp "chmod +x /tmp/npm"
If git messed up:
# Uncommit last commit (keep changes)
git reset --soft HEAD~1
# Discard all uncommitted changes
git restore .
# Return to clean state
git checkout navidocs-cloud-coordination
git pull origin navidocs-cloud-coordination
Cost Estimates (Token Economics)
| Approach | Model | Tokens | Cost | Time |
|---|---|---|---|---|
| Sonnet reads agents.md | Sonnet | ~50K | $1.50 | 5 min |
| Sonnet does all tasks | Sonnet | ~200K | $6.00 | 2 hours |
| Haiku swarm (5 agents) | Haiku | ~80K | $0.80 | 1.5 hours |
| Mixed (Sonnet coordinates, Haiku works) | Both | ~120K | $2.40 | 1.5 hours |
Recommendation: Mixed approach (what previous Claude did)
- Sonnet: Read docs, coordinate swarm, synthesize results
- Haiku: Environment setup, deployment, testing, documentation
Budget remaining: Unknown (user didn't specify limit)
User Preferences (Communication Style)
User likes:
- ✅ Extreme clarity in documentation
- ✅ Step-by-step breakdowns
- ✅ Proactive status updates
- ✅ S² autonomous coordination (less human in the loop)
- ✅ IF.TTT compliance (citations for everything)
User dislikes:
- ❌ Vague "it should work" claims
- ❌ Not updating agents.md
- ❌ Assumptions without verification
- ❌ Making user the bottleneck (waiting for relay)
Tone:
- Professional but not overly formal
- Technical depth appreciated
- Don't use emojis unless user does first (noted in agents.md)
- Focus on facts > validation
Known Issues (What's Broken or Weird)
-
StackCP npm/npx don't work out of the box
- Issue: Scripts use
#!/usr/bin/env nodewhich fails on noexec home directory - Fix: Wrapper scripts in /tmp/ (documented in STACKCP_README.md)
- Status: Not yet created (YOUR TASK)
- Issue: Scripts use
-
Git branch diverged
- Issue: Local branch 4 commits ahead, remote 14 commits ahead
- Impact: May cause merge conflicts
- Fix:
git fetch && git mergeor rebase if comfortable - Status: Needs attention
-
Feature selector exportAgentTasks() not implemented
- Issue: Button exists but function incomplete
- Impact: Can't generate agent task JSON
- Fix: Add function (YOUR TASK #1)
- Status: High priority
-
No Redis Cloud credentials yet
- Issue: Backend needs Redis for BullMQ job queue
- Impact: Can't start background workers
- Fix: User will provide tomorrow OR use free tier signup
- Status: Blocked until user wakes OR we self-serve
-
No Google Cloud Vision API key
- Issue: OCR fallback unavailable
- Impact: Handwriting OCR won't work
- Fix: User providing tomorrow OR test Claude Code CLI only
- Status: Workaround available (/tmp/claude for OCR)
Emergency Contacts (When Completely Stuck)
Technical Issues:
- Check agents.md (lines 11-1217)
- Check SESSION_HANDOVER_2025-11-13.md (full context)
- Check STACKCP_README.md via SSH
- Check S² narration (methodology)
- Create ESCALATION.md file for user
Can't Find Something?
# Find files
find /home/setup -name "*.md" -type f | grep -i navidocs
# Search content
grep -r "feature selector" /home/setup/navidocs/
# Check git history
git log --all --oneline --grep="feature"
StackCP Access Issues?
# Test SSH
ssh -v stackcp "echo Connection successful"
# Check SSH config
cat ~/.ssh/config | grep -A 5 stackcp
# Check key permissions
ls -la ~/.ssh/icw_stackcp_ed25519
# Should be: -rw------- (600)
Git Issues?
# Check remotes
git remote -v
# Check branch
git branch -a
# Check status
git status
# See what's uncommitted
git diff
Things That Will Save You Hours
-
Read agents.md lines 11-60 FIRST (Update protocol)
- Saves: 2 hours of confusion
- Impact: Critical for next agent
-
Use Haiku swarm for parallel tasks
- Saves: 3-4 hours of sequential work
- Impact: 5× speedup
-
Verify after EVERY step
- Saves: 1-2 hours of debugging at the end
- Impact: Catch issues early
-
Test locally before deploying to StackCP
- Saves: 15-30 minutes per iteration
- Impact: Faster feedback loop
-
Keep agents.md open in second window
- Saves: Constant file switching
- Impact: Reference while working
Questions Future Claudes Always Ask
Q: Where's the NaviDocs codebase?
A: /home/setup/navidocs (local), will be ~/navidocs-app on StackCP
Q: What's the feature selector URL? A: https://digital-lab.ca/navidocs/builder/
Q: How do I SSH to StackCP?
A: ssh stackcp (alias configured) or see agents.md credentials section
Q: What's /tmp vs ~/ on StackCP? A: /tmp = executables only (persistent!), ~/ = everything else (noexec but persistent)
Q: Do I need to read all 94 intelligence files? A: NO! Feature summary in agents.md lines 1133-1217. Full intelligence if you need deep details.
Q: How much budget do I have? A: Unknown. User didn't specify. Use Haiku for cost efficiency. Estimate $2-5 for remaining work.
Q: What if I mess up agents.md?
A: git diff agents.md to see changes, git restore agents.md to undo, commit often
Q: Can I skip the S² narration? A: Technically yes, but you'll miss 3,563× coordination speedup methodology. Worth 15 min read.
End of New Session Startup Guide
Next agent: Read this file → Read agents.md → Execute Task 1 → Deploy Task 2 (Haiku swarm) → Update agents.md → Commit
User: Wake up to working feature selector + summary document
🚤 Generated with Claude Code
IF.TTT Citation: if://decision/new-session-startup-guide-2025-11-13