https://github.com/splitbrain/wdid
what did i do? claude remembers
https://github.com/splitbrain/wdid
claude-code vibe-coded
Last synced: 8 days ago
JSON representation
what did i do? claude remembers
- Host: GitHub
- URL: https://github.com/splitbrain/wdid
- Owner: splitbrain
- Created: 2026-05-04T21:36:57.000Z (about 2 months ago)
- Default Branch: main
- Last Pushed: 2026-05-05T06:57:59.000Z (about 2 months ago)
- Last Synced: 2026-05-05T08:34:51.724Z (about 2 months ago)
- Topics: claude-code, vibe-coded
- Language: Python
- Homepage:
- Size: 10.7 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
README
# wdid — what did I do?
A small CLI that reports your Claude Code activity for a given day, derived
from the JSONL session logs under `~/.claude/projects/`.
## Usage
```
./wdid # today
./wdid yesterday
./wdid 2026-04-30
./wdid --no-summarize # skip Haiku-based descriptions
```
Output:
```
START END DUR PROJECT DESC
-----------------------------------------------------------------------------------
09:12 10:04 00:48 api Add rate limiting middleware and integration tests
10:20 11:35 01:11 webapp Migrate user settings page to new component library
13:30 14:02 00:31 scripts Investigate slow nightly backup job
-----------------------------------------------------------------------------------
total 02:30 (3 sessions)
```
## How it figures out engagement
A single Claude Code session log can span many real working sessions
(`claude --resume` reattaches to old logs hours or days later). `wdid`
walks each log, sorts events by timestamp, and splits into engagement
sessions wherever the gap between consecutive events exceeds 30 minutes.
Engagement time per session is the sum of consecutive in-session gaps
(naturally capped at 30 min by the splitting rule), plus a 3-minute
trailing read buffer when the session ends on an assistant message — the
logs have no close/exit signal, so we credit roughly the same idle window
that triggers Claude's own `away_summary`.
Subagent logs (`.../subagents/...`) are skipped — they overlap with their
parent session and would double-count time.
## Session descriptions
The `DESC` column comes from, in order of preference:
1. A Haiku-generated summary (cached at `~/.cache/wdid/titles-YYYY-MM-DD.json`).
The summary is built from your prompts, the tool actions Claude took
(file edits, shell commands), and the last few assistant replies, so it
tends to reflect what was actually done rather than just the opening
prompt.
2. Claude's own `ai-title` entry, if Haiku is disabled or fails.
3. The first non-slash user prompt.
## Requirements
- Python 3.9+ (standard library only)
- Optional: `ANTHROPIC_API_KEY` environment variable for Haiku-based
summaries. Without it, `wdid` falls back to `ai-title` / first prompt.
The summarizer uses `claude-haiku-4-5-20251001` via the Anthropic
Messages API, called directly through `urllib`. Each title is cached, so
subsequent runs make zero API calls.