Warm Winter SDK v0.3.0

Native computational consultation

Python now provides consultation_capabilities(), consult_queue() and consultation_feedback(). TypeScript/JavaScript provides consultationCapabilities(), consultQueue() and consultationFeedback().

The helpers use the existing consultation API and authenticated client transports. Queue inputs include rates, units, caller source references, usefulness forecast and optional task/run lineage. Responses preserve the full server envelope: assumptions, provenance, unknown uncertainty, resource limits, advisory status and overload null markers. Usefulness feedback remains self-reported and does not independently verify task success or grant execution authority.

An explicitly granted consultation-only key is required. Existing Gate/observer keys gain no new access, and the SDK does not mint grants or install automatic hooks. Server admission, revocation and rate-limit controls remain authoritative. Existing Gate methods remain unchanged; no runtime dependencies are added. The backend-coupled contextual extensions are not included.

Validation and source

The local candidate wheel and source archive each passed all 11 Python SDK tests. The local npm tarball passed legacy logging checks, all 11 consultation contracts and packaged declaration checks. Tests used synthetic transport without live service calls. The publishing workflow rebuilds distributions from this tag; local archive hashes are verification evidence, not uploaded-byte receipts.

Reviewed candidate: c39ea02 (PR #5).
Integrated source: 34ea086; identical Git tree 91e631a621f7d49ca6f944486747e9627e0f76f5.

Python: pip install warmwinter==0.3.0
JavaScript: npm install warmwinter@0.3.0

Warm Winter SDK v0.2.0

What changed

• Branded, prompt-blind verdict logging in Python and TypeScript, with opt-out and custom logger support.
• Python strong-label outcome resolution by commit SHA for CI and deploy health checks.
• Python local decisions now honor per-cell stakes ceilings.
• Package-level privacy, version, strong-label, and stakes tests run before trusted publishing.

Both SDKs remain zero-dependency at runtime.