VPS Cleanup

Systematic approach to cleaning up VPS disk space — analyze files, categorize by importance, safely delete temporary/old data without losing critical files.

When to Use

• User asks "apa yang gak penting?" or "what can I delete?"
• Disk space running low (> 80% used)
• Home directory cluttered with temporary files
• After completing projects (cleanup workspace)

Workflow

Phase 1: Analyze Current State

# Check disk usage
df -h /

# Analyze home directory
cd /home/ubuntu
du -sh * 2>/dev/null | sort -h

# Check large directories
du -h --max-depth=1 | sort -h | tail -20

Phase 2: Categorize Files

For each file/directory, assess:

Phase 3: Build Decision Matrix

Create analysis table:

files = {
    'file.zip': {
        'size': '135M',
        'type': 'backup',
        'keep': True,
        'reason': 'Important backup'
    },
    'temp-workspace/': {
        'size': '538M',
        'type': 'workspace',
        'keep': False,
        'reason': 'Temporary analysis, reports done'
    },
    # ... more files
}

Calculate:

• Total size
• Deletable size
• Keep size
• Percentage reclaimable

Phase 4: Present Analysis

Show user:

1. Summary stats (total, keep, delete, %)
2. Keep list (sorted by size, with reasons)
3. Delete list (sorted by size, with reasons)
4. Recommendations (phased approach)

Format:

✅ KEEP (IMPORTANT)
  135M  backup.zip        Important backup
  2.4M  snap/             System directory

❌ CAN DELETE (NOT IMPORTANT)
  538M  temp-workspace/   Temporary, work done
  1.2M  diagrams/         PNGs generated

Phase 5: Confirm with User

CRITICAL: Always confirm before deleting, especially for:

• Large directories (> 100 MB)
• Project directories (might have uncommitted work)
• Anything user-created (not system files)

Ask:

• "Mau gue delete sekarang?" (Want me to delete now?)
• "Ada yang mau keep?" (Anything you want to keep?)

Phase 6: Safe Deletion

# Delete directories
rm -rf ~/temp-workspace/
rm -rf ~/old-project/

# Delete files
rm ~/old-archive.tar.gz
rm ~/report.md

# Verify
ls -lh ~/
df -h /

Safety checks:

• Use rm -rf only for confirmed directories
• Delete one category at a time (easier to undo if mistake)
• Verify disk space freed after deletion

User Preference Patterns

For "Delete All Unimportant"

When user says "delete semua yang gak penting" or "delete all unimportant":

DO:

• ✅ Analyze and categorize first
• ✅ Present full list with reasons
• ✅ Ask about ambiguous items (projects, large files)
• ✅ Delete after confirmation

DON'T:

• ❌ Delete immediately without analysis
• ❌ Skip showing what will be deleted
• ❌ Delete backups or system files
• ❌ Delete active projects without asking

For Safety Confirmation

When presenting cleanup plan, user may respond:

• "Aman di bersihkan?" (Safe to clean?) → User wants reassurance
• ✅ Confirm: "100% AMAN" + list what won't be deleted
• ✅ Explain worst-case scenario (e.g., "cache can be regenerated")
• ✅ Emphasize zero data loss
• ✅ Then ask: "Gas cleanup?" (Ready to proceed?)

• "Ya" (Yes) → Proceed immediately
• ✅ Execute cleanup commands
• ✅ Show progress for each step
• ✅ Verify results (before/after disk usage)
• ✅ Celebrate success: "✅ Cleanup complete! Freed XMB"

User's decision-making pattern:

1. Asks for analysis → wants to see what's unused
2. Asks "aman?" → wants safety confirmation
3. Says "ya" → ready to execute
4. Expects immediate action after confirmation (no further questions)

For Ambiguous Items

Projects (e.g., oh-my-china/):

🟡 AMBIGUOUS: oh-my-china/ (340 MB)
   Type: Project
   Question: "Masih aktif develop atau cuma testing?"
   
   If active → KEEP
   If testing → DELETE

Generated outputs:

🟡 OPTIONAL: diagram.png (4 MB)
   Type: Output
   Question: "Masih perlu atau gak?"
   
   If needed → KEEP
   If done → DELETE

Common File Types

Temporary Workspaces (DELETE)

Patterns:

• *-analysis/, *-test/, *-workspace/
• Cloned repos for one-time analysis
• Temporary diagram folders

Indicators:

• Reports already generated
• Work completed
• No uncommitted changes

Example:

hermes-analysis/  # 538 MB
├─ 10 cloned repos
├─ ANALYSIS_REPORT.md (done)
└─ install script (used)

→ DELETE (reports done, repos installed elsewhere)

Old Archives + Extracted (DELETE)

Pattern: Both .tar.gz AND extracted folder exist

Example:

gmail_research/              # 100 KB (extracted)
gmail_research_2026-05-01.tar.gz  # 20 KB (archive)

→ DELETE BOTH (duplicate, old data)

Temporary Reports (DELETE)

Patterns:

• *_REPORT.md, *_AUDIT.md
• Generated for one-time review
• Info already applied

Example:

VPS_SECURITY_AUDIT_REPORT.md      # 12 KB
VPS_SECURITY_HARDENING_REPORT.md  # 12 KB

→ DELETE (hardening done, info applied)

Source Files (OPTIONAL)

Pattern: Source exists, output generated

Example:

ai-promo.html     # 16 KB (source)
ai-promo.png      # 4 MB (output)

→ DELETE SOURCE (output is final)

Backups (KEEP)

Patterns:

• *-backup-*.zip, *.tar.gz with dates
• Named with "backup" or "archive"

Example:

hermes-backup-2026-05-03.zip  # 135 MB

→ KEEP (important backup!)

System Directories (KEEP)

Patterns:

• snap/, .cache/, .local/, .config/
• Hidden directories (.hermes/, .ssh/)

Example:

snap/        # 2.4 MB
.hermes/     # 2.6 GB

→ KEEP (system/application data)

Resource Evaluation for Tool Installation

CRITICAL: Before recommending ANY tool installation, evaluate resource cost against VPS constraints.

VPS Resource Constraints (User: ryzen, VPS: 2GB RAM, 40GB disk)

User tolerance:

• ❌ ZERO tolerance for heavy tools
• ❌ Disk > 500 MB = instant rejection
• ❌ RAM > 100 MB = ask first or find lightweight alternative
• ✅ Prefer API/cloud services over local installation

Evaluation criteria:

Example rejection (Scrapling, session 2026-05-12):

User: "Cek Scrapling bagus gak?"
Analysis:
- Disk: 650-900 MB (7.0 MB library + browser dependencies)
- RAM: 200-400 MB per session (browser automation)
- VPS: 2GB RAM, 40GB disk
- Features: Powerful (adaptive parsing, Cloudflare bypass, spider framework)

Comparison with TinyFish:
- TinyFish: 0 MB disk, 0 MB RAM (cloud API)
- Scrapling: 650-900 MB disk, 200-400 MB RAM (local)

User response: "gede banget makan emori dan ram"
Decision: SKIP ❌ → Recommend TinyFish instead

Lesson: User has ZERO tolerance for heavy tools.
Always check resource cost BEFORE recommending.
When tool is heavy, immediately suggest lightweight alternative (API/cloud).

User's exact rejection pattern:

• "gede banget makan emori dan ram" = instant rejection, no discussion
• When user says this, STOP immediately — don't offer workarounds, don't suggest "minimal install", don't explain benefits
• Acknowledge rejection, move on to lightweight alternatives
• This phrase signals: resource cost is deal-breaker, not negotiable

Alternative strategies:

1. Cloud API (Browserbase, Browser Use Cloud) — 0 MB local cost
2. Lightweight alternatives (requests + BeautifulSoup vs Scrapling)
3. Docker (isolated, but still heavy)
4. Skip feature (if not critical)

Common heavy tools and their costs:

enowxai v1.2.0 specific:

• Requires Chrome/Chromium (300-400 MB disk)
• Worker pool spawns multiple Chrome instances
• Risk: OOM on 2GB RAM VPS with concurrent workers
• Mitigation: Limit concurrent workers to 2-3 max

Pitfalls

0. Recommending Deletion Without Thorough Analysis

Problem: Suggesting file cleanup without checking references, purpose, or user preference.

Solution:

• ALWAYS analyze before recommending deletion:
• Check if files are referenced elsewhere (grep -r filename)
• Verify file purpose (marketing, docs, tools, assets)
• Consider total size impact (is cleanup worth it?)
• Assess user's attachment to files
• Present analysis, let user decide:
• Show: file list + sizes + purposes + references
• Explain: why each file might be kept or deleted
• Ask: "Mau gue hapus atau keep semua?" (Delete or keep all?)
• User preference pattern:
• "jangan hapus kalo berguna" = user wants to keep useful files
• "brarti semua aman?" = user wants confirmation, not assumptions
• When in doubt, present analysis and ASK

Example (from session 2026-05-12):

User: "hapus file2 yang gak perlu"
❌ WRONG: Immediately recommend deleting marketing files
✅ RIGHT: Analyze all files, check references, present findings:
  - Marketing files (7.5 KB): Not referenced, but useful for sharing
  - Deep docs (42 KB): Redundant with README, but valuable for advanced users
  - Total size: 284 KB (tiny, not worth cleanup)
  - Recommendation: Keep all (every file has purpose, size is negligible)

1. Deleting Backups

Problem: User says "delete all", you delete backups too.

Solution:

• Always categorize backups as KEEP
• Highlight backups in analysis
• Confirm separately if user wants to delete backups

2. Deleting Active Projects

Problem: Large project directory looks like temporary workspace.

Solution:

• Check for .git/ (active repo)
• Check last modified date (recent = active)
• Ask user if unsure

3. Not Showing Reasons

Problem: User doesn't understand why something is kept/deleted.

Solution:

• Always include reason in analysis
• Be specific: "Temporary workspace, reports done" not just "temporary"

4. Deleting Without Confirmation

Problem: User says "delete unimportant", you delete immediately.

Solution:

• Always show analysis first
• Get explicit confirmation
• Especially for large items (> 100 MB)

5. Not Verifying Disk Space Freed

Problem: Delete files but don't show impact.

Solution:

• Show disk usage before/after
• Confirm space was actually freed
• Celebrate success: "✅ Freed 884 MB!"

Verification

After cleanup:

# Check disk space
df -h /

# Check home directory
ls -lh ~/

# Verify critical files still exist
ls ~/.hermes/
ls ~/snap/

Success criteria:

• ✅ Disk space freed (check df -h)
• ✅ Critical files intact (backups, system dirs)
• ✅ User satisfied with cleanup
• ✅ No accidental deletions

Advanced: Cleanup ~/.hermes/

CAUTION: Only clean ~/.hermes/ if user explicitly requests or disk critically low.

# Check size
du -sh ~/.hermes/

# Analyze subdirectories
du -h ~/.hermes/ --max-depth=1 | sort -h

# Safe to clean:
- Old sessions: hermes sessions prune --older-than 30d
- Cache: rm -rf ~/.hermes/.cache/
- Logs: rm ~/.hermes/*.log

# DO NOT delete:
- config.yaml
- memories/
- skills/
- sessions.db (unless pruning)

Hermes Ecosystem Audit: See references/hermes-ecosystem-audit.md for:

• Systematic audit methodology (4-phase process)
• Common cleanup targets (state snapshots, WAL files, logs, caches)
• Unused vs. inactive distinction (CRITICAL — bundled plugins are inactive, not unused)
• Audit report template
• Maintenance schedule (weekly/monthly tasks)
• Key learnings from 2026-05-05 audit (78MB cleanup, zero data loss)

Key insight from audit: "Installed" ≠ "Active" ≠ "Consuming resources"

• Bundled code (plugins, toolsets) has zero cost until executed
• Only active processes consume RAM
• Only written files consume disk (code is static)
• Example: 8 memory provider plugins installed = 1.2MB disk, 0MB RAM (inactive, not unused)

Examples

Example 1: Simple Cleanup (540 MB freed)

Analysis:
- hermes-analysis/ (538 MB) → Temporary workspace
- drawio-test/ (1.2 MB) → Temporary diagrams
- old reports (24 KB) → Info applied

Decision: DELETE ALL (safe, no data loss)

Result: 540 MB freed ✅

Example 2: Cleanup with Confirmation (884 MB freed)

Analysis:
- hermes-analysis/ (538 MB) → DELETE
- oh-my-china/ (340 MB) → ASK USER (project?)
- ai-promo.png (4 MB) → ASK USER (still need?)

User: "delete semua yang gak penting"

Confirm:
- oh-my-china/ → "Masih develop?" → "Cuma testing" → DELETE
- ai-promo.png → "Masih perlu?" → "Gak" → DELETE

Result: 884 MB freed ✅

References

• df man page: https://man7.org/linux/man-pages/man1/df.1.html
• du man page: https://man7.org/linux/man-pages/man1/du.1.html
• Safe deletion practices: https://www.gnu.org/software/coreutils/manual/html_node/rm-invocation.html
• Scrapling vs TinyFish comparison: references/scrapling-vs-tinyfish-comparison.md (tool evaluation for VPS constraints)