Daily Review AI Transcripts

Daily Review — AI Transcripts

Gathers yesterday's AI session transcripts from Claude Code JSONL files and the Conductor app database, then saves a structured summary to sessions/YYYY-MM-DD-ai-sessions.md.

Trigger

Invoked as part of /daily-review or independently via /daily-review-ai-transcripts.

Context

• Working directory: ~/Projects/ikigai
• Session storage: sessions/
• Date: Use yesterday's date (YYYY-MM-DD)

Process

Skip check

Skip if sessions/YYYY-MM-DD-ai-sessions.md already exists for yesterday. Report "already exists, skipping" and stop.

Source A: Claude Code JSONL files

1. Find all JSONL files modified yesterday:

find ~/.claude/projects -maxdepth 2 -name "*.jsonl" -not -path "*/subagents/*" -type f | while read f; do
  mod=$(stat -f '%Sm' -t '%Y-%m-%d' "$f")
  if [ "$mod" = "YYYY-MM-DD" ]; then
    created=$(stat -f '%SB' -t '%H:%M' "$f")
    modified=$(stat -f '%Sm' -t '%H:%M' "$f")
    size=$(stat -f '%z' "$f")
    proj=$(echo "$f" | sed 's|.*/\.claude/projects/||' | sed 's|/[^/]*$||')
    echo "$proj|$created|$modified|$size|$f"
  fi
done | sort

2. Group by project. For the top 3 projects by total size, extract human messages:

import json
with open('FILE') as f:
    for line in f:
        try:
            obj = json.loads(line)
            if obj.get('type') == 'human':
                msg = obj.get('message', {}).get('content', '')
                if isinstance(msg, list):
                    for c in msg:
                        if isinstance(c, dict) and c.get('type') == 'text':
                            t = c['text'].strip()
                            if t and not t.startswith('<'): print(t[:500])
                elif isinstance(msg, str) and msg.strip() and not msg.startswith('<'):
                    print(msg[:500])
        except: pass

Source B: Conductor app database

3. Query the Conductor SQLite database for yesterday's sessions:

DB="$HOME/Library/Application Support/com.conductor.app/conductor.db"
sqlite3 "$DB" "
  SELECT s.title, w.directory_name, sm.role, sm.content, sm.created_at
  FROM session_messages sm
  JOIN sessions s ON sm.session_id = s.id
  LEFT JOIN workspaces w ON s.workspace_id = w.id
  WHERE date(sm.created_at) = 'YYYY-MM-DD'
    AND sm.role = 'user'
  ORDER BY sm.created_at
"

• Group by workspace (maps to project)
• Extract user prompts from content column (JSON format)
• Conductor sessions have titles and workspace context for richer summaries

Save results

4. Save to sessions/YYYY-MM-DD-ai-sessions.md with:
   • Time spent per project (created->modified timestamps, session count, total size)
   • Insights and learnings from the top sessions
   • Full path to every raw transcript (JSONL paths for Claude Code sessions)
   • Conductor session titles and workspace names for Conductor sessions

Output

The file path of the saved AI sessions file, or a note that it was skipped.