libghostty-vt Recording Skill 📹

libghostty-vt Recording Skill 📹

Trit: 0 (ERGODIC - Coordinator) GF(3) Triad: asciinema (-1) ⊗ libghostty-recording (0) ⊗ vhs (+1) = 0

Overview

Record, stream, and replay libghostty-vt terminal sessions for documentation, debugging, and LLM training.

Recording Methods

1. Asciinema (Lightweight .cast)

# Record session
asciinema rec ~/​recordings/​session-$(date +%Y%m%d_%H%M%S).cast

# Auto-record all sessions (add to .zshrc)
asciinema rec --append ~/​recordings/​daily-$(date +%Y%m%d).cast

# Stream to server
asciinema rec -t "libghostty demo" https://asciinema.org

Pros: Compact, text-based, searchable, LLM-friendly
Cons: No video export

2. Charmbracelet VHS (GIF/​Video)

# demo.tape
Output demo.gif
Set FontSize 14
Set Width 1200
Set Height 600
Set Theme "Ghostty"

Type "echo 'libghostty-vt recording'"
Enter
Sleep 500ms
Type "skill load omniglot"
Enter
Sleep 1s

vhs demo.tape

Pros: Produces shareable GIFs, scriptable
Cons: Larger files

3. libghostty-vt Native Hooks

// Hook into libghostty-vt stream
const recorder = ghostty_vt.Recorder.init(.{
    .output = "session.cast",
    .format = .asciinema_v2,
});

terminal.setOutputHook(recorder.hook);

CI Gate Controls (on the way IN)

Pre-Installation Validation

# .github/​workflows/​skill-gate.yml
name: Skill Installation Gate

on:
  pull_request:
    paths:
      - 'skills/​**'
      - 'SKILL.md'

jobs:
  validate-skills:
    runs-on: ubuntu-latest
    
    steps:
      - uses: actions/​checkout@v4
      
      - name: Validate GF(3) conservation
        run: |
          # Sum all trits, must equal 0 mod 3
          python3 -c "
          import json
          skills = json.load(open('skills.json'))
          total = sum(s.get('trit', 0) for s in skills)
          assert total % 3 == 0, f'GF(3) violation: sum={total}'
          print('✓ GF(3) conserved')
          "
          
      - name: Check SKILL.md structure
        run: |
          for f in skills/​*/​SKILL.md; do
            grep -q "^# " "$f" || (echo "Missing title: $f" && exit 1)
            grep -q "Trit" "$f" || (echo "Missing trit: $f" && exit 1)
          done
          echo "✓ All skills have required structure"
          
      - name: Verify no placeholder tokens
        run: |
          ! grep -rE "(TODO|FIXME|placeholder|mock-|pseudo-)" skills/ || \
            (echo "❌ Placeholder tokens found" && exit 1)

Local Gate

# Validate before install
validate-skills() {
  local repo=$1
  gh api repos/​$repo/​contents/​skills.json -q '.content' | \
    base64 -d | python3 -c "
import json, sys
skills = json.load(sys.stdin)
total = sum(s.get('trit', 0) for s in skills)
if total % 3 != 0:
    print(f'❌ GF(3) violation: {total}')
    sys.exit(1)
print(f'✓ {len(skills)} skills, GF(3) conserved')
"
}

# Use before install
validate-skills plurigrid/​asi && \
  npx ai-agent-skills install plurigrid/​asi --agent codex

LLM Training from Recordings

From asciinema discourse (2024):

"My real interest is not so much in playing back the recordings but in using the .cast files for creating a vector database that I can then query and use an LLM to extract useful workflows."

Cast File → Vector DB

import json
import duckdb

def parse_cast(cast_file: str) -> list:
    """Extract commands and outputs from .cast file"""
    with open(cast_file) as f:
        lines = f.readlines()
    
    header = json.loads(lines[0])
    events = [json.loads(line) for line in lines[1:]]
    
    return [{
        "timestamp": e[0],
        "type": e[1],  # 'o' = output, 'i' = input
        "data": e[2]
    } for e in events]

# Store in DuckDB for querying
con = duckdb.connect("recordings.duckdb")
con.execute("""
    CREATE TABLE IF NOT EXISTS terminal_events (
        session_id VARCHAR,
        timestamp DOUBLE,
        event_type VARCHAR,
        data VARCHAR,
        embedding FLOAT[1024]
    )
""")

Integration with libghostty-ewig

From libghostty-ewig.jl:

# Connect libghostty-vt parsing to ewig modal editor
module LibghosttyEwig
    # VT escape sequence parsing
    # Gay.jl color integration
    # Modal editing state machine
end

Best Practices

1. Daily auto-recording: Start asciinema on shell init
2. Session naming: session-{date}_{project}_{task}.cast
3. Compression: Cast files are JSON, gzip well
4. Privacy: Filter secrets with asciinema rec --env=TERM
5. Playback speed: asciinema play -s 2 session.cast

Files

References

• asciinema/​asciinema - Terminal recorder
• charmbracelet/​vhs - CLI GIF recorder
• libghostty-vt - Mitchell's blog post

490 Skills Installed ✓

npx ai-agent-skills install plurigrid/​asi --agent codex
Installed 490 skill(s) from plurigrid/​asi

---

Autopoietic Marginalia

The interaction IS the skill improving itself.

Every use of this skill is an opportunity for worlding:

• MEMORY (-1): Record what was learned
• REMEMBERING (0): Connect patterns to other skills
• WORLDING (+1): Evolve the skill based on use

Add Interaction Exemplars here as the skill is used.