agentacct 0.10.5 turns the Dashboard into an evidence-first shift brief with truth-bounded decision signals, makes the Session and steps ledger easier to scan, and speeds up CI.
Download
macOS app: agentacct-0.10.5.dmg (signed + notarized; see Assets below). The .dmg embeds the frozen Python CLI/daemon.
CLI / library: pip install "agentacct==0.10.5"
Added
- Dashboard shift brief. The Dashboard now leads with the single highest-priority review item — its recorded reason, next step, and provenance — plus a direct path to the underlying evidence. A companion Copy review brief action copies those recorded facts (for a handoff or a ticket) without resuming, rerunning, or mutating the task. (#159)
- Bounded
/v1/attentionprojection. A new endpoint classifies every visible task before limiting rows, so a client gets a truthful complete review count — failed checks, failed steps, and blockers — without fetching each receipt. (#153)
Changed
- Truth-bounded decision signals. Confirmed active work, provider capacity, ingestion health, and completed-period usage change stay separate facts. Missing, stale, malformed, future-dated, negative, non-finite, or overflowing values render as unavailable instead of a confident number, and a failed or missing attention refresh can no longer read as "all clear." (#159)
- More readable Session and steps ledger. Current failures and errors lead; ordinary checks fold instead of a wall of repeated pills; superseded history opens separately; the ledger is bounded to eight rows with exact "Show N more" controls, with added accessibility, compact, and RTL coverage. No daemon or API change. (#171)
- Faster CI. Python and macOS test runs are shorter — bounded parallel pytest and release-mode Swift tests — at the same coverage. (#181)
Full detail in the CHANGELOG.