Harbor is a project-partitioned research and monitoring system.

Phase P4 complete — refresh & error recovery: manual fetch-now, snapshot history inline, staleness + fetch-error signals on the overview. Release cut as v0.3.0-alpha (VPS-deployment baseline). Next: VPS deployment (local-on-VPS first, then decide on public reach). Feature bolts (P5 cite-back, cross-project dedup, P6 automated search) deferred until deploy is accepted.

Harbor currently contains:

• accepted A0 product and architecture baseline
• runtime bootstrap and local operator surface
• persistence foundation and Postgres baseline
• manual operator research workflow:
  • project
  • source / project-source
  • search campaign
  • review queue
  • search run
  • search result candidate
  • candidate/review/source promotion flow
  • duplicate guards
  • workflow summary and lineage
• operator web shell under /operator/...
• OpenAI adapter baseline
• project dry-run surface
• persisted OpenAI dry-run log history
• chat surface under /chat
• persisted chat sessions and chat turns
• multi-turn chat context hardening
• chat turn inspection and compare readability surfaces
• chat composer / instructions UX split
• chat instructions preset/default UX
• source-grounded chat with enriched metadata and citation extraction
• handbook context injection in chat prompt
• operator action: propose source from chat
• operator action: draft handbook entry from chat
• source-review workflow available inside the operator web shell
• handbook version history surface in the operator web shell
• operator JS and CSS extracted to src/harbor/static/ (operator.js, chat.js, operator.css)
• automation task registry + API (observability baseline for automation-triggered work)
• side-channel observer pattern — failures are recorded even when the request transaction rolls back
• collapsible operator section-cards with per-operator localStorage persistence
• automation task log panel on project-detail (kind / trigger / status / timestamps / result-or-error)

• v0.3.0-alpha — current baseline: snapshot pipeline + scheduler + content activation + refresh/error recovery; VPS-deployment-ready with SQLite
• v0.2.0-alpha — scheduler + snapshot baseline (T6/T7 + U1–U6 operator UX primitives)
• v0.1.0-alpha — manual operator flow baseline

See CHANGELOG.md for version history and docs/RELEASE_NOTES_v0_3_0_alpha.md for current release notes.

Canonical local path:

• C:\projekte\Harbor

1. docs/MASTERPLAN.md
2. docs/PROJECT_STATE.md
3. docs/STRATEGY_ROADMAP_v0_1.md
4. docs/WORKING_AGREEMENT.md
5. docs/VALIDATION_PROTOCOL.md
6. docs/_handoff/HANDOFF_2026-04-12_P4_to_next.md

.\.venv\Scripts\Activate.ps1

python .\tools\task_runner.py show-settings
python .\tools\task_runner.py show-db-settings
python .\tools\task_runner.py quality-gates

python .\tools\task_runner.py smoke-project-slice
python .\tools\task_runner.py smoke-source-slice
python .\tools\task_runner.py smoke-search-campaign-slice
python .\tools\task_runner.py smoke-search-run-slice
python .\tools\task_runner.py smoke-search-result-candidate-slice
python .\tools\task_runner.py smoke-review-queue-slice
python .\tools\task_runner.py smoke-candidate-review-promotion-slice
python .\tools\task_runner.py smoke-review-queue-source-promotion-slice
python .\tools\task_runner.py smoke-promotion-duplicate-guard-slice
python .\tools\task_runner.py smoke-workflow-summary-slice
python .\tools\task_runner.py smoke-operator-web-shell-slice
python .\tools\task_runner.py smoke-openai-adapter-slice
python .\tools\task_runner.py smoke-openai-project-dry-run-slice
python .\tools\task_runner.py smoke-openai-chat-session-slice
python .\tools\task_runner.py smoke-chat-surface-slice

• /healthz
• /runtime
• /db/status
• /operator/projects
• /operator/projects/{project_id}
• /chat
• /api/v1/openai/runtime
• /api/v1/openai/probe
• /api/v1/openai/projects/{project_id}/dry-run
• /api/v1/openai/projects/{project_id}/dry-run-logs
• /api/v1/openai/projects/{project_id}/chat-sessions
• /api/v1/openai/projects/{project_id}/chat-sessions/{chat_session_id}/turns
• /api/v1/openai/projects/{project_id}/chat-turns
• /api/v1/openai/projects/{project_id}/propose-source
• /api/v1/openai/projects/{project_id}/draft-handbook
• /api/v1/projects/{project_id}/automation-tasks
• /api/v1/automation-tasks/{automation_task_id}

• 291 tests, 95.55% coverage (70% gate enforced)
• 14 Alembic migrations (linear chain, integrity-tested)
• 14 ORM models
• typed domain exceptions, middleware exception handlers
• request-scoped transaction middleware (commit-on-success/rollback-on-error)
• side-channel observer pattern for observability across rollback
• structured request logging with request-id propagation
• expanded Ruff lint rules: E, F, I, B, UP, SIM, PIE, LOG, RUF
• 24 permanent engineering manifest rules