thaitype Claude Session Log
Replay a Claude Code session as a human-readable transcript or structured JSON. Shows assistant text, tool calls, and tool results for every turn.
Commands
# Show log for a specific session
python3 .claude/skills/claude-session-log/scripts/session_log.py <session-id>
# Show log for the most recent session
python3 .claude/skills/claude-session-log/scripts/session_log.py --latest
# JSON output (always full content, no truncation)
python3 .claude/skills/claude-session-log/scripts/session_log.py --latest --json
# Show only tool calls and results (skip pure text)
python3 .claude/skills/claude-session-log/scripts/session_log.py --latest --tools-only
# Disable truncation on tool results (default: 500 chars)
python3 .claude/skills/claude-session-log/scripts/session_log.py --latest --full
# Use custom projects directory
python3 .claude/skills/claude-session-log/scripts/session_log.py --latest --dir /path/to/projects
Output
Human-readable (default):
Session: 82c946c8-1eb8-4c70-9414-813fc0a278e4
Date: 2026-03-28 14:58
Model: claude-opus-4-6
Turns: 5
--- user (14:58:04) ---
Fix the bug in parser.py
--- assistant (14:58:06) ---
Let me read the file first.
> Read scripts/parser.py
[2847 chars, truncated]
def parse(input):
tokens = input.split()
...
> Edit scripts/parser.py
old: if t.startswith("--"):
new: if t.startswith("--") or t.startswith("-"):
Fixed the parser to handle single-dash flags.
--- user (14:58:12) ---
< tool_result (Read): [ok]
< tool_result (Edit): [ok]
--- assistant (14:58:15) ---
Done. The parser now handles both --flag and -f syntax.
JSON mode (--json):
{
"session_id": "82c946c8-...",
"model": "claude-opus-4-6",
"turns": 5,
"messages": [
{
"role": "user",
"timestamp": "2026-03-28T14:58:04Z",
"content": [{"type": "text", "text": "Fix the bug..."}]
},
{
"role": "assistant",
"timestamp": "2026-03-28T14:58:06Z",
"content": [
{"type": "text", "text": "Let me read the file first."},
{"type": "tool_use", "name": "Read", "input": {"file_path": "scripts/parser.py"}}
]
}
]
}