Manifold v0.2.2

Manifold v0.2.2

Manifold v0.2.2 separates long-running Brain document ingestion from the
ordinary dependency timeout. A production release write-back showed that the
pinned Brain service could finish extraction successfully in just over 30
seconds while Manifold's shared 30-second HTTP client had already recorded the
durable job as partially_ready or failed.

Brain document ingestion now uses MANIFOLD_BRAIN_INGEST_TIMEOUT, which
defaults to 2 minutes. Health, search, and other interactive dependency calls
retain the existing short timeout. A regression test proves that a slow ingest
can complete through the dedicated client even when the ordinary client would
time out.

This release includes the Brain search-limit and multiline integration-oracle
fixes from v0.2.1 and every protocol, recovery, folder-deletion, and deployment
improvement from v0.2.0.

After the server reports 0.2.2 and protocol revision 2, update the globally
installed skill:

npx skills update manifold --global

Manifold v0.2.1

Manifold v0.2.1

Manifold v0.2.1 fixes context retrieval against the pinned INITE Brain
contract. The service intentionally asks each backend for extra candidates
before reciprocal-rank fusion, but context retrieval requested 200 Brain
candidates while Brain v0.8.1 accepts at most 100. Brain rejected that request
with HTTP 400, so otherwise healthy context responses reported Brain as a
degraded dependency.

The Brain adapter now clamps its backend-specific candidate request to 100.
This preserves the public Manifold limit and rank-fusion behavior while keeping
the upstream request valid. A contract test protects the boundary. The
integration assertion was also corrected to recognize non-empty dependency
arrays in pretty-printed, multiline JSON, eliminating the false-green that hid
the production failure.

This patch release includes all v0.2.0 features and operational fixes. Deploy
the server first, verify /api/v1/meta reports 0.2.1 and protocol revision 2,
then update the globally installed skill:

npx skills update manifold --global

Manifold v0.2.0

Manifold v0.2.0

Manifold v0.2.0 is the first versioned operational release of the
self-hosted knowledge and memory layer. Earlier deployments reported 0.1.0
or dev independently of their source commit; this release makes version,
protocol revision, and exact build SHA verifiable through the running service.

Highlights

• Adds discovery-first remember, atomic nested folder_path creation, and
  slash-aware folder/document tree selectors for external agents.
• Adds public GET /api/v1/meta with API major, protocol revision, exact build
  commit, and stable feature names.
• Keeps base search, context, and tree usable across a legacy server/client
  skew by omitting zero-value optional fields.
• Refuses unsupported filtered retrieval and every unverified legacy mutation
  with server_incompatible and exit 23 instead of silently changing intent.
• Repairs INITE Brain scoped SurrealDB session recovery after a database
  restart; integration tests keep Brain running while restarting SurrealDB and
  require semantic retrieval to recover without degradation.
• Reconciles deletion of empty OpenViking folders that contain derived semantic
  entries, including bounded retries while OpenViking releases parent path
  locks after nested cleanup.
• Adds exact-SHA production deployment after green main CI, deep authenticated
  smoke checks, serialized host deployment, and rollback without deleting
  .env or named volumes.
• Publishes native skill clients and the server container for amd64 and
  arm64.

Upgrade order

Deploy the v0.2.0 server first, verify /api/v1/meta reports protocol revision
2 and every dependency is ready, then update the globally installed skill:

npx skills update manifold --global

For a first installation:

npx skills add iamwavecut/Manifold \
  --skill manifold \
  --global

The installed skill requires only MANIFOLD_URL and MANIFOLD_API_KEY at
runtime; Node.js is used by the installer, not by the bundled client.