Why this exists

Productivity tooling splits into two camps that never talk to each other. Task boards record intent: what you planned to do and what stage it reached. Time trackers record raw clock time: how long some window was open. Neither answers the question that actually matters at the end of a day, which is whether the work you planned is where your attention really went. The gap between the two is where drift lives and no amount of manual logging survives long enough to expose it.

Locus closes that gap by putting both halves in one native window. Work moves through four fixed stages (Plan, Execute, Check, Done) while the runtime reads foreground focus directly from the Win32 APIs, polling into the same local SQLite database as the tasks. The record is automatic and honest: idle gaps are subtracted rather than counted, system processes are filtered out and application names come from real PE version info instead of a hardcoded mapping.

The third ingredient is the Claude Code integration. Hook scripts route each meaningful tool call onto the board as it happens: edits land at Execute, test runs at Check and a successful command moves its card to Done, while trivial commands are dropped. An AI coding session normally evaporates as scrollback; here it leaves a structured trail on the same board as the work you planned yourself.

It is built the way a small Windows tool should be: a single Go binary with no CGO, installed per-user with one PowerShell script, auto-starting on login, with no telemetry, no cloud and no accounts. Locus grew out of two earlier tools (CommandDeck's staged session model and focus-reader's background focus monitor) and is the native Go consolidation of both.

Full reasoning at crankthecode.com.