A keyboard-native agentacct tui rebuilt to mirror the macOS app — now with a receipt → sessions & steps drill-down — an honest "Inactive" task state that never overstates completion, and a receipt-first README refresh.
Install
macOS app (signed & notarized): agentacct-0.10.6.dmg — download, open, drag to Applications. The .app embeds the frozen CLI (importer, daemon, MCP server), so it ships the full 0.10.6 backend.
CLI (PyPI):
pipx install "agentacct==0.10.6" # or: uv tool install "agentacct==0.10.6"
What's new
Added
agentacct tui, rebuilt to mirror the macOS app: a four-tab, keyboard-native shell — Dashboard / Work / Usage / Sources — with light and dark themes and one shared evidence/decision/cost vocabulary. Opening a receipt drills into its sessions & steps — the checks timeline behind the verdict, with a currently-failing check kept visible under Needs attention, the passing checks below, and the files it touched. Plus a?help overlay, a/filter and status tabs on Work, an Evidence Sources pane, and aTlight/dark toggle. (#185)- An honest "Inactive" task state. A task shown "in progress" is downgraded to an evidence-bounded Inactive — never a completion claim — only when it has open steps, nothing recorded as finished, and a genuinely newer session kept the store working elsewhere for 48h past that session's start. Computed from stored timestamps, never the wall clock, so silence or clock skew can only keep a task active, never falsely retire it; provenance is
inferred(never green) and it stays out of the review queue. (#184)
Changed
- The TUI redesign is a presentation-layer rewrite over the same local event log the CLI and macOS app read, so the surfaces cannot disagree. (#185)
- Receipt-first README refresh for the shipped UI. (#183)
Fixed
- Pin
textualto>=8,<9so a future Textual major release cannot break the installed TUI. (#185)
Full changelog: https://github.com/mikehasa/agentacct/blob/v0.10.6/CHANGELOG.md