lost-rob0t Analyze ActivityWatch data
Goal
Turn ActivityWatch window, AFK, web, and editor buckets into evidence-backed workflow findings without treating raw activity as a productivity score.
Workflow
- Discover available buckets from the local ActivityWatch API or user-provided export. Confirm the time range and bucket types before making claims.
- Normalize app classes, domains, titles, and editor buffers into broad categories. Redact credentials, tokens, private URLs, personal prompts, document contents, and identifying path segments.
- Measure time totals, active intervals, transitions, repeated sequences, interruption clusters, and missing-data boundaries. Distinguish observed duration from event count.
- Report the evidence, uncertainty, and a short list of workflow changes. Compare periods only when collection coverage and bucket semantics match.
Rules
- Use the local server by default (
http://127.0.0.1:5600); never upload ActivityWatch data or silently query a remote service. - Do not infer intent, health, employment performance, or moral value from app names or idle time. Label interpretations as hypotheses.
- Prefer aggregate categories and counts. Show raw titles or URLs only when the user explicitly requests them and they are safe to disclose.
- Suggestions may target aliases, Emacs commands, Qtile rules, project jumps, or reminders, but implementation requires user approval.