ai-efficiencylisted
Install: claude install-skill xSubahx/ai-efficiency
# AI Efficiency
A Claude subscription window opens with the first message and resets five hours later.
Left alone, windows start wherever the first prompt of a stretch happens to land, so a
heavy afternoon can be spent waiting on a window that opened late in the morning. This
skill finds the cycle that fits the user's real rhythm and keeps it running.
It never raises a limit. It changes *when* windows start, so more of the user's working
hours sit inside a fresh window. The weekly limit is untouched and still applies.
## 1. Analyse
Run the analyser. It reads only timestamps, models and token counts from
`~/.claude/projects`, and nothing leaves the machine. It takes about half a minute.
```sh
python3 "<skill dir>/scripts/analyze.py" # readable report
python3 "<skill dir>/scripts/analyze.py" --json # the same, for you to reason over
```
Options: `--days 31` (history to replay), `--root DIR` (repeatable, e.g. transcripts
copied from a second machine), `--step 30` (minutes between candidate start times).
If it reports fewer than five active days, say so and stop: there is not enough history
for a schedule worth trusting.
## 2. Recommend
Tell the user, in plain words and in this order:
1. **When they work** - the two or three busiest stretches from the hourly profile.
2. **The recommended cycle** - the wake-up times and the quiet stretch
(`recommended.pings`, `recommended.quiet`).
3. **What it buys** - windows their work touched today versus with the cy