rewatch 0.2.0a1

analytics over your claude code session transcripts

rewatch

analytics over your claude code session transcripts.

claude code keeps a jsonl transcript of every session under ~/.claude/projects/. rewatch separates what you actually typed from harness noise (tool results, slash-command expansions, subagent sidechains) and tells you things like:

• your most-repeated literal prompts (everyone has a "make the PR description not suck")
• how often you open with no / wait / stop / actually
• where your prompts, interrupts, and output tokens go, per project
• which tools claude leans on, and what hours you really work

install

uv tool install rewatch

usage

rewatch report            # the whole picture, human-formatted
rewatch prompts --pretty  # most-repeated literal prompts
rewatch hours --pretty    # prompts by local hour of day
rewatch overview --days 7 # recent only

machine-readable by default — lists are ndjson, single results are one json object — so output pipes straight into jq or an agent:

rewatch projects | jq -r '.project'

note: claude code prunes local transcripts (30 days by default, see cleanupPeriodDays), so rewatch sees a rolling window, not all time.

library

from rewatch import load_sessions, repeated_prompts

sessions = load_sessions(days=30)
for text, count in repeated_prompts(sessions, top=10):
    print(count, text)