analyse-bottlenecklisted
Install: claude install-skill jindraregal/analyse-bottleneck
# analyse-bottleneck
A script does the analysis; you read its compact output and add judgement. You never read raw transcripts.
## Default flow: one chart
`${CLAUDE_SKILL_DIR}` is the directory that contains this file (Claude Code fills it in; if you see it unfilled, use that directory). `OUT` is where the user wants reports (default `~/.local/share/analyse-bottleneck/reports/`).
1. Run the view (same scope options as below: `--project`, `--all`, `--session`, `--since`, `--until`, `--sysstat`, `--jobs`, `--dora-json`):
```bash
python3 ${CLAUDE_SKILL_DIR}/scripts/analyze.py view --project my-app --since 7d --out OUT --name my-app
```
It writes `OUT/my-app.view.json` (the data), `OUT/my-app.html`, `OUT/my-app.brief.md` (<= 60 lines) and, when a deep dive triggers, `OUT/my-app.deepdive-<id>.md`.
The page is fully built by the script, the same way every time: the title, one plain sentence, four small stats with icons (possible saving, recommendations, period, sessions) and "How to read this"; a donut of agent time plus a hatched "Waiting for you" slice, with agent time and the waiting in its centre. It opens on "All slices" ("Do this first" across slices, then any deep dive); picking a slice shows its "Do this" (coded rules, AI notes) and "What exactly" (items with 2-5 word names from `scripts/rules/describe.json` or the agent's own call descriptions); picking it again, the centre, "Show all" or Esc goes back to All. Every list is the same expandable row, and e