CoolFace
Modelpublic

smlflg/claude-insights-timeline

sourceHugging Faceupdated 10d agoView on Hugging Face
0likes
Model Card

claude-insights-timeline

Standalone Python script that builds filterable Markdown timelines from Claude Code /insights data — without LLM calls.

/insights overwrites report.html on every run. This tool reads the cached per-session JSONs (facets/ + session-meta/) and produces chronological, filterable timeline snapshots.

Usage

bash
python3 insights-timeline.py [filters]

Filters (all optional, combinable via AND)

FlagDescription
--since YYYY-MM-DDSessions from date (inclusive)
--until YYYY-MM-DDSessions until date (inclusive)
--friction-min NOnly sessions with total friction >= N
--session-type TYPEexploration, implementation, debugging, ...
--project SUBSTRINGSubstring match on project path
--output PATHOverride output file path
--stdoutAlso print to stdout

Examples

bash
# Full timeline
python3 insights-timeline.py

# April sessions only
python3 insights-timeline.py --since 2026-04-01

# Problem sessions
python3 insights-timeline.py --friction-min 3

# Specific project
python3 insights-timeline.py --project Sidecar

# Combined: high-friction Sidecar sessions in March
python3 insights-timeline.py --project Sidecar --friction-min 2 --since 2026-03-01 --until 2026-03-31

Output

Markdown file written to ~/.claude/usage-data/queries/timeline-YYYY-MM-DD-HHMM.md (default) or --output path.

Each session entry includes: timestamp, duration, project, session type, outcome, summary, friction detail, top tools, and first prompt snippet.

Data Source

Reads from ~/.claude/usage-data/:

  • —facets/<uuid>.json — LLM-generated per-session analysis (cached by /insights)
  • —session-meta/<uuid>.json — harness metrics (tool counts, tokens, timestamps)

No API calls. No dependencies beyond Python 3.10+ stdlib.