Startempire-Wire/focusa

Keep AI coding agents on mission.
Local-first proof, Workpoints, Evidence, and continuation for long-running coding agents.
Focusa is the local-first proof and continuity layer for AI coding agents. Current source version: v0.9.209.
When a coding session gets long, context compacts, the mission drifts, proof gets buried, or another agent takes over, Focusa preserves the work as a proof-backed **Workpoint** with linked **Evidence** and a **next safe action**. The next agent should not have to guess from transcript memory.
Current release architecture
- **Exact scope and worktrees:** canonical cognition binds ScopeRef/ProjectRootKey + WorkstreamId + ContinuityId and applicable AttachmentKey; project root or continuity alone is insufficient. Verified worktrees remain working subpaths, not accidental new projects.
- **Daemon-native Silent Sessions:** durable background runs support observation, steering, pause/resume/restart, approvals, idempotency, and receipts.
- **Governed work loop and recovery:** one writer, canonical checkpoints, proactive compaction, cache-safe context, and automatic rollover after bounded transport exhaustion.
- **Mission Canvas and Work Rail:** scoped Work Surfaces, CRIST interviews, workspace artifacts, UIAI browser context, live refresh, connectors, software/domain projections, and adaptive generated UI.
- **Customer lifecycle:** idempotent install/repair, trusted Linux/macOS/Windows release assets, OTA/update rollback, and uninstall with user data preserved unless purge is explicit.
- **Release provider map:** temporary Linux/GitHub-self-hosted, Windows/AppVeyor, and macOS/Codemagic routing—and the one-change-set return to GitHub—is defined in docs/178-focusa-temporary-ci-provider-parity-and-github-restoration-spec.md.
- **Agent-ready by construction:** every Focusa Pi tool has a runtime contract, generated machine schema, per-tool document, skill routing, and a progressively disclosed runbook.
- **Pi reference harness:** Pi + the Focusa Pi extension is the default/reference Focusa-aware harness integration. That role is fundamental to the reference agent experience but does not move cognitive authority out of Focusa daemon/core or make non-Pi harnesses second-class.
- **Focusa Desktop presenter:** Focusa Desktop is a governed presenter over the same Focusa authority and operation guards. In supported full Veragensia Agent Computer profiles it is the default human work/cognition surface; it does not become a second reducer, state store, or product authority.
- **Voice / Conversation primitive:** Doc 08 Expression Engine remains the semantic expression owner; Spec 181 defines full-duplex spoken interaction, speaker/agent attribution, transcript correction lineage, interruption, group conversation and a local-first Conversation Ledger while preserving the rule that conversation is not memory and voice is not authority.
- **Project Foreman:** Spec 182 defines the Foreman as the persistent Workstream-scoped projection of project intelligence across models, sessions, Desktop, mobile, voice and workers—not another chatbot or memory store.
- **Radar:** Spec 183 defines proactive scoped observation, deduplicated Episodes, Signal economics and attention routing while preserving observation != authority and Spec 139 ownership of runtime presence/placement.
- **Ambient Operator:** Spec 184 projects the same governed system through phone/earbuds/wearables for bounded presence, wake/conversation, meetings, Radar attention, Foreman/Wirebot routing and private/offline sync without making the device a second brain.
Trajectory ladder: **HLT** (High-Level Trajectory) → **MLG** (Mid-Level Goal) → **STG** (Short-Term Goal) → **Waypoints** (concrete progress markers). Workpoint remains immediate action authority. The operator has authority; agents actively offer HLT-aligned Waypoints, STGs, and MLGs without silently changing the root goal.
One cohesive agent journey