BigBrain v1.0.0

Added

• Added a Brain-local page schema registry with MCP tools to inspect,
  bootstrap, register, update, and delete schema definitions, plus health
  checks for schema compliance and first-class page summaries.
• Added first-class frameworks, articles, and videos collections,
  multi-image page attachments, article capture, and dedicated page, search,
  Substack, podcast, YouTube, email-management, and ingestion skills.
• Added the Stream dashboard with source and collection filters, attached-image
  browsing, complete source counts, and bounded infinite pagination.
• Added Comprehension and Keep in Touch playbooks with private overlay state,
  bounded reviews, resumable cursors, and dashboard surfaces.
• Added one shared verified RSS and Substack ingestion ledger with atomic
  claims, retryable outcomes, explicit partial captures, scoped read-back
  evidence, and per-feed retained-coverage bounds.

Changed

• Search now embeds and returns complete ranked page evidence for harness-side
  synthesis, overlaps lexical and semantic retrieval, and applies bounded
  reranking to complete page content.
• New structured timeline entries require source provenance. Task and page
  mutations now share the canonical page writer and reject storage-oriented
  timeline prose.
• Canonical pages now support a first-class summary stored in frontmatter and
  rendered separately from their detailed body. Health reports missing
  summaries as maintenance findings.
• Background event workers now recover orphaned claims and queue backpressure,
  isolate native computer use, and expose only the MCP and browser capabilities
  required by ingestion work.
• Focused graph navigation, labels, source flow, node controls, lifecycle
  handling, and page reading views have been substantially refined.

Fixed

• Fixed hosted browser OAuth onboarding and remote dashboard capability
  handling.
• Fixed graph focus after rotation, transient particle teardown, label overlap,
  and focused-network recovery.
• Fixed RSS retry visibility, verified completion checks, and Granola route
  verification so queue or worker prose cannot falsely prove ingestion.

Removed

• Removed the MCP natural-language query answer tool. Use search for
  complete evidence and let the calling harness synthesize the answer. The CLI
  query compatibility command remains retrieval-only.
• Removed the standalone conversation-ingest skill. Conversation capture is now
  part of the general ingestion workflow.

Agent update actions

• This is a major upgrade. Before updating, back up the Brain directory,
  runtime database, ~/.config/bigbrain, and the machine-local RSS/Substack
  ledger. Existing schema-v1 databases are migrated to embedding schema v2 on
  first use; rollback to an older runtime is not supported after migration.
• Source installs: update to v1.0.0, run npm install, npm link, and
  npm run build:dashboard, then restart the desktop, dashboard, MCP, event
  ingestor, and RSS services.
• Existing Brain configs do not automatically gain new schema directories.
  Merge frameworks, articles, and videos into schema_dirs without
  overwriting local configuration. Bootstrap the page schema registry, review
  it, and commit page-schemas.yaml before enforcing custom schemas.
• Run a full bigbrain sync --json with the deployment's embedding credentials
  configured so complete-page embeddings are generated. Then run
  bigbrain health --json and resolve missing page summaries, strict-schema
  findings, and legacy task fields. Only exact blank legacy assignees or
  source strings may be normalized automatically to empty arrays; review all
  other values manually.
• Refresh active bundled skills and remove any active
  bigbrain-conversation-ingest copy. Review the new email-ingest,
  Keep in Touch, and Substack-reconciliation automations before activating
  them, and preserve local schedules, targets, and customizations.
• Server and container deployments must persist the Brain, database, OAuth
  state, and the shared ledger path configured by
  BIGBRAIN_SUBSTACK_LEDGER_PATH. Never reconstruct verified article outcomes
  from feed cursors or prose logs.
• MCP clients must replace query calls with search. Structured timeline
  writers must provide complete provenance with event_id, source_type,
  source_label, and source_message for every new entry.
• Verify /live, /ready, the reported runtime version, authenticated
  tools/list, search, page schema tools, RSS/Substack ledger status,
  Comprehension and Keep in Touch playbooks, bigbrain sync --json,
  bigbrain health --json, and npm test after deployment.

Verification

• Local-data compatibility audit passed with the required major-version and
  migration actions above. Legacy timeline entries remain readable, task enum
  values are unchanged, and shared ledger migrations preserve existing rows.
• Full test suite passed: 678 tests passed, 7 skipped because
  TEST_DATABASE_URL is not configured, and 0 failed.
• Dashboard production build and standalone MCP bundle verification passed as
  release gates before tagging.